Prof. Dr.-Ing. Ribana Roscher

Junior Professor of Remote Sensing
Contact:
Email: ribana.roscher@nulluni-bonn.de
Tel: +49 - 228 - 73 - 27 16
Fax: +49 - 228 - 73 - 27 12
Office: Nussallee 15, 2. OG, room 2.001
Address:
University of Bonn
Photogrammetry, IGG
Nussallee 15
53115 Bonn
Google Scholar
LinkedIn
ResearchGate
Download CV

Research Interests

  • Landcover classification methods
  • Sparse representation for large scale remote sensing data
  • Incremental/Sequential learning
  • Feature/Representation learning

Short CV

Ribana Roscher is a Junior Professor of Remote Sensing at the University of Bonn. Before working in Bonn, she was researcher at The Fields Institute for Research in Mathematical Sciences, Toronto, Canada, Julius Kühn-Institut (Institute for Grapevine Breeding) in Siebeldingen, Humboldt Innovation in Berlin and Freie Universität Berlin. Ribana Roscher finished her PhD thesis in 2012 entitled “Sequential learning using incremental import vector machines for semantic segmentation” supervised by Wolfgang Förstner at University of Bonn. In her research, she aims at the development of pattern recognition methods, which are particularly designed for the analysis of large scale remote sensing data. She specifically focusses on efficient classification methods, techniques for sophisticated feature learning and the integration of prior knowledge such as spatial and temporal information. A central idea in her research is to develop methods which ensure a high discrimination power and at the same time model the underlying structure of the data, since such methods are a prerequisite for the automatic analysis of earth observation data.

Current Research

Self-taught learning by sparse representation

Self-taught learning with archetypal analysis

Self-taught learning (STL) has become a promising paradigm to exploit large amounts of unlabeled data for feature learning and classification. It utilizes both labeled and unlabeled data without the requirement that both sets have to share the same distribution and the same land use and land cover classes. In this context, archetypal analysis is used to extract the most valuable unlabeled data which is used in the STL framework to learn an efficient and discriminative feature representation. In our research the approach is used for land cover classification of multispectral images.
Research partners: Remote Sensing and Geoinfomatics Working Group, FU Berlin

Detected diseases on a sugar beet plant leaf

Sparse representation for anomaly detection

Here the benefit of using various structured dictionaries, such as topographic dictionaries, is analyzed for sparse representation (SR). The approach is used for the detection of Cercospora leaf spot disease symptoms on sugar beet plants in hyperspectral images.
Furthermore, sparse representation can be used to build sparse graphs for spectral clustering with highly distinctive clusters. The approach can be used for identifying clusters for change detection in remote sensing satellite images and anomaly detection in close range hyperspectral images.
Research partners: INRES-Pflanzenkrankheiten und Pflanzenschutz, University of Bonn

Spatio-temporal modelling of altimetric waveforms

Spatio-Temporal Altimeter Waveform Retracking

Satellite radar altimetry is one of the most powerful techniques for measuring sea surface height variations, with applications ranging from operational oceanography to climate research. In this research we develop a novel spatio-temporal altimetry retracking technique which elements are the integration of information from spatially and temporally neighboring waveforms, a sub-waveform detection, where relevant sub-waveforms are separated from corrupted or non-relevant parts, and the identification the final best set of sea surfaces heights from multiple likely heights using a shortest-path algorithm.
Research partners: Satellite Geodesy Group, University of Bonn

Further Activities

Awards

  • 2016: IEEE GRSS Symposium Prize Paper Award for the IGARRS paper ‘Landcover classification with self-taught learning on archetypal dictionaries’
  • 2009: Turbo-Preis of Society for Geodesy, Geoinformation and Land Management (DVW)

Teaching Information

Talks

  • Unsupervised and Self-taught Learning for Remote Sensing Image Analysis, Keynote at Workshop on Environmental Informatic Challenges, 2017
  • Sparse Representation-based Analysis of Hyperspectral Remote Sensing Data, International Workshop on Perspectives on High Dimensional Data, 2017
  • Unsupervised learning for remote sensing data, SPLIT Remote Sensing Summer School, 2017
  • Der objektive Blick von oben – Wie der Computer unsere Umwelt sieht, Tag der Geodäsie, 2017
  • Maschinelles Lernen für Anwendungen in der Landwirtschaft, Keynote at 23. Computer-Bildanalyse-Workshop, 2017
  • Unsupervised and self-taught learning for remote sensing image interpretation, Group for Research in Decision Analysis (GERAD) Seminar Series, HEC Montreal, 2017
  • Pattern recognition and machine learning for remote sensing, Colloquium, Zentrum für Fernerkundung der Landoberfläche (ZFL, Center for Remote Sensing of Land Surfaces), University of Bonn, 2017
  • Unüberwachtes Lernen für die Interpretation von Fernerkundungsbilder, Inaugural lecture, University of Bonn, 2016
  • Detection of disease symptoms on hyperspectral 3D plant models, ISPRS Congress, 2016
  • On the benefit of topographic dictionaries for detecting disease symptoms on hyperspectral 3D plant models, Workshop on Hyperspectral Image and Signal Processing, 2016
  • Discriminative and reconstructive methods for classification of remote sensing images, Microsoft Research Computational Statistics and Machine Learning Seminar Series, University College London, 2015
  • Discriminative and reconstructive methods for classification of remote sensing images, Group for Research in Decision Analysis (GERAD) Seminar Series, HEC Montreal, 2015
  • Landcover classification with self-taught learning on archetypal dictionaries, IEEE International Geoscience and Remote Sensing Symposium (IGARSS), GRSS Symposium Prize Paper Award 2015
  • Spatio-temporal altimeter waveform retracking via sparse representation and conditional random fields, IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2015
  • Shapelet-based sparse image representation for landcover classification of hyperspectral data, IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS), 2014
  • Superpixel-based classification of hyperspectral data using sparse representation and conditional random fields, IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2014
  • Logistische Regression für die semantische Segmentierung in der Photogrammetrie und Fernerkundung, Deutsche Gesellschaft für Photogrammetrie und Fernerkundung (DGPF), 2011
  • Kernel discriminative random fields for land cover classification, IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS), 2010

Publications

2017

  • L. Drees and R. Roscher, “Archetypal Analysis for Sparse Representation-based Hyperspectral Sup-Pixel Quantification,” ISPRS Annals of Photogrammetry, Remote Sensing & Spatial Information Sciences, vol. 4, 2017.
    [BibTeX]
    @Article{drees2017archetypal,
    Title = {Archetypal Analysis for Sparse Representation-based Hyperspectral Sup-Pixel Quantification},
    Author = {Drees, Lukas and Roscher, Ribana},
    Journal = {ISPRS Annals of Photogrammetry, Remote Sensing \& Spatial Information Sciences},
    Year = {2017},
    Volume = {4}
    }

  • J. Rosentreter, R. Hagensieker, A. Okujeni, R. Roscher, and B. Waske, “Sub-pixel mapping of urban areas using EnMAP data and multioutput support vector regression,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017. doi:10.1109/JSTARS.2017.2652726
    [BibTeX]
    @Article{Rosentreter2017Subpixel,
    Title = {Sub-pixel mapping of urban areas using EnMAP data and multioutput support vector regression},
    Author = {Rosentreter, J. and Hagensieker, R. and Okujeni, A. and Roscher, R. and Waske, B.},
    Journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
    Year = {2017},
    Note = {to appear},
    Doi = {10.1109/JSTARS.2017.2652726},
    Owner = {ribana},
    Timestamp = {2017.01.08}
    }

2016

  • B. Franke, J. Plante, R. Roscher, A. Lee, C. Smyth, A. Hatefi, F. Chen, E. Gil, A. Schwing, A. Selvitella, M. M. Hoffman, R. Grosse, D. Hendricks, and N. Reid, “Statistical Inference, Learning and Models in Big Data,” International Statistical Review, 2016.
    [BibTeX] [PDF]
    Big data provides big opportunities for statistical inference, but perhaps even bigger challenges, often related to differences in volume, variety, velocity, and veracity of information when compared to smaller carefully collected datasets. From January to June, 2015, the Canadian Institute of Statistical Sciences organized a thematic program on Statistical Inference, Learning and Models in Big Data. This paper arose from presentations and discussions that took place during the thematic program.

    @Article{Franke2016BigData,
    Title = {Statistical Inference, Learning and Models in Big Data},
    Author = {Franke, Beate and Plante, Jean-Fran\c{c}ois and Roscher, Ribana and Lee, Annie and Smyth, Cathal and Hatefi, Armin and Chen, Fuqi and Gil, Einat and Schwing, Alex and Selvitella, Alessandro and Hoffman, Michael M. and Grosse, Roger and Hendricks, Dieter and Reid, Nancy},
    Journal = {International Statistical Review},
    Year = {2016},
    Note = {to appear},
    Abstract = {Big data provides big opportunities for statistical inference, but perhaps even bigger challenges, often related to differences in volume, variety, velocity, and veracity of information when compared to smaller carefully collected datasets. From January to June, 2015, the Canadian Institute of Statistical Sciences organized a thematic program on Statistical Inference, Learning and Models in Big Data. This paper arose from presentations and discussions that took place during the thematic program.},
    Owner = {ribana},
    Timestamp = {2016.03.01},
    Url = {http://onlinelibrary.wiley.com/doi/10.1111/insr.12176/full}
    }

  • B. Mack, R. Roscher, S. Stenzel, H. Feilhauer, S. Schmidtlein, and B. Waske, “Mapping raised bogs with an iterative one-class classification approach,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 120, pp. 53-64, 2016. doi:http://dx.doi.org/10.1016/j.isprsjprs.2016.07.008
    [BibTeX] [PDF]
    Abstract Land use and land cover maps are one of the most commonly used remote sensing products. In many applications the user only requires a map of one particular class of interest, e.g. a specific vegetation type or an invasive species. One-class classifiers are appealing alternatives to common supervised classifiers because they can be trained with labeled training data of the class of interest only. However, training an accurate one-class classification (OCC) model is challenging, particularly when facing a large image, a small class and few training samples. To tackle these problems we propose an iterative \{OCC\} approach. The presented approach uses a biased Support Vector Machine as core classifier. In an iterative pre-classification step a large part of the pixels not belonging to the class of interest is classified. The remaining data is classified by a final classifier with a novel model and threshold selection approach. The specific objective of our study is the classification of raised bogs in a study site in southeast Germany, using multi-seasonal RapidEye data and a small number of training sample. Results demonstrate that the iterative \{OCC\} outperforms other state of the art one-class classifiers and approaches for model selection. The study highlights the potential of the proposed approach for an efficient and improved mapping of small classes such as raised bogs. Overall the proposed approach constitutes a feasible approach and useful modification of a regular one-class classifier.

    @Article{Mack2016Raised,
    Title = {Mapping raised bogs with an iterative one-class classification approach },
    Author = {Mack, Benjamin and Roscher, Ribana and Stenzel, Stefanie and Feilhauer, Hannes and Schmidtlein, Sebastian and Waske, Bj{\"o}rn},
    Journal = {{ISPRS} Journal of Photogrammetry and Remote Sensing},
    Year = {2016},
    Pages = {53 - 64},
    Volume = {120},
    Abstract = {Abstract Land use and land cover maps are one of the most commonly used remote sensing products. In many applications the user only requires a map of one particular class of interest, e.g. a specific vegetation type or an invasive species. One-class classifiers are appealing alternatives to common supervised classifiers because they can be trained with labeled training data of the class of interest only. However, training an accurate one-class classification (OCC) model is challenging, particularly when facing a large image, a small class and few training samples. To tackle these problems we propose an iterative \{OCC\} approach. The presented approach uses a biased Support Vector Machine as core classifier. In an iterative pre-classification step a large part of the pixels not belonging to the class of interest is classified. The remaining data is classified by a final classifier with a novel model and threshold selection approach. The specific objective of our study is the classification of raised bogs in a study site in southeast Germany, using multi-seasonal RapidEye data and a small number of training sample. Results demonstrate that the iterative \{OCC\} outperforms other state of the art one-class classifiers and approaches for model selection. The study highlights the potential of the proposed approach for an efficient and improved mapping of small classes such as raised bogs. Overall the proposed approach constitutes a feasible approach and useful modification of a regular one-class classifier. },
    Doi = {http://dx.doi.org/10.1016/j.isprsjprs.2016.07.008},
    ISSN = {0924-2716},
    Keywords = {Remote sensing},
    Url = {http://www.sciencedirect.com/science/article/pii/S0924271616302180}
    }

  • R. Roscher, J. Behmann, A. -K. Mahlein, J. Dupuis, H. Kuhlmann, and L. Plümer, “Detection of Disease Symptoms on Hyperspectral 3D Plant Models,” in ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences , 2016, pp. 89-96.
    [BibTeX]
    We analyze the benefit of combining hyperspectral images information with 3D geometry information for the detection of Cercospora leaf spot disease symptoms on sugar beet plants. Besides commonly used one-class Support Vector Machines, we utilize an unsupervised sparse representation-based approach with group sparsity prior. Geometry information is incorporated by representing each sample of interest with an inclination-sorted dictionary, which can be seen as an 1D topographic dictionary. We compare this approach with a sparse representation based approach without geometry information and One-Class Support Vector Machines. One-Class Support Vector Machines are applied to hyperspectral data without geometry information as well as to hyperspectral images with additional pixelwise inclination information. Our results show a gain in accuracy when using geometry information beside spectral information regardless of the used approach. However, both methods have different demands on the data when applied to new test data sets. One-Class Support Vector Machines require full inclination information on test and training data whereas the topographic dictionary approach only need spectral information for reconstruction of test data once the dictionary is build by spectra with inclination.

    @InProceedings{Roscher2016detection,
    Title = {Detection of Disease Symptoms on Hyperspectral {3D} Plant Models},
    Author = {Roscher, R. and Behmann, J. and Mahlein, A.-K. and Dupuis, J. and Kuhlmann, H. and Pl{\"u}mer, L.},
    Booktitle = {ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences},
    Year = {2016},
    Pages = {89--96},
    Abstract = {We analyze the benefit of combining hyperspectral images information with 3D geometry information for the detection of Cercospora leaf spot disease symptoms on sugar beet plants. Besides commonly used one-class Support Vector Machines, we utilize an unsupervised sparse representation-based approach with group sparsity prior. Geometry information is incorporated by representing each sample of interest with an inclination-sorted dictionary, which can be seen as an 1D topographic dictionary. We compare this approach with a sparse representation based approach without geometry information and One-Class Support Vector Machines. One-Class Support Vector Machines are applied to hyperspectral data without geometry information as well as to hyperspectral images with additional pixelwise inclination information. Our results show a gain in accuracy when using geometry information beside spectral information regardless of the used approach. However, both methods have different demands on the data when applied to new test data sets. One-Class Support Vector Machines require full inclination information on test and training data whereas the topographic dictionary approach only need spectral information for reconstruction of test data once the dictionary is build by spectra with inclination.}
    }

  • R. Roscher, J. Behmann, A. -K. Mahlein, and L. Plümer, “On the Benefit of Topographic Dictionaries for Detecting Disease Symptoms on Hyperspectral 3D Plant Models,” in Workshop on Hyperspectral Image and Signal Processing , 2016.
    [BibTeX]
    We analyze the benefit of using topographic dictionaries for a sparse representation (SR) approach for the detection of Cercospora leaf spot disease symptoms on sugar beet plants. Topographic dictionaries are an arranged set of basis elements in which neighbored dictionary elements tend to cause similar activations in the SR approach. In this paper, the dictionary is obtained from samples of a healthy plant and partly build in a topographic way by using hyperspectral as well as geometry information, i.e. depth and inclination. It turns out that hyperspectral signals of leafs show a typical structure depending on depth and inclination and thus, both influences can be disentangled in our approach. Rare signals which do not fit into this model, e.g. leaf veins, are also captured in the dictionary in a non-topographic way. A reconstruction error index is used as indicator, in which disease symptoms can be distinguished from healthy plant regions.nThe advantage of the presented approach is that full spectral and geometry information is needed only once to built the dictionary, whereas the sparse reconstruction is done solely on hyperspectral information.

    @InProceedings{Roscher2016Topographic,
    Title = {On the Benefit of Topographic Dictionaries for Detecting Disease Symptoms on Hyperspectral 3D Plant Models},
    Author = {Roscher, R. and Behmann, J. and Mahlein, A.-K. and Pl{\"u}mer, L.},
    Booktitle = {Workshop on Hyperspectral Image and Signal Processing},
    Year = {2016},
    Abstract = {We analyze the benefit of using topographic dictionaries for a sparse representation (SR) approach for the detection of Cercospora leaf spot disease symptoms on sugar beet plants. Topographic dictionaries are an arranged set of basis elements in which neighbored dictionary elements tend to cause similar activations in the SR approach. In this paper, the dictionary is obtained from samples of a healthy plant and partly build in a topographic way by using hyperspectral as well as geometry information, i.e. depth and inclination. It turns out that hyperspectral signals of leafs show a typical structure depending on depth and inclination and thus, both influences can be disentangled in our approach. Rare signals which do not fit into this model, e.g. leaf veins, are also captured in the dictionary in a non-topographic way. A reconstruction error index is used as indicator, in which disease symptoms can be distinguished from healthy plant regions.nThe advantage of the presented approach is that full spectral and geometry information is needed only once to built the dictionary, whereas the sparse reconstruction is done solely on hyperspectral information.},
    Owner = {ribana},
    Timestamp = {2016.06.20}
    }

  • R. Roscher, S. Wenzel, and B. Waske, “Discriminative Archetypal Self-taught Learning for Multispectral Landcover Classification,” in Proc. of Pattern Recogniton in Remote Sensing 2016 (PRRS), Workshop at ICPR; to appear in IEEE Xplore , 2016.
    [BibTeX] [PDF]
    Self-taught learning (STL) has become a promising paradigm to exploit unlabeled data for classification. The most commonly used approach to self-taught learning is sparse representation, in which it is assumed that each sample can be represented by a weighted linear combination of elements of a unlabeled dictionary. This paper proposes discriminative archetypal self-taught learning for the application of landcover classification, in which unlabeled discriminative archetypal samples are selected to build a powerful dictionary. Our main contribution is to present an approach which utilizes reversible jump Markov chain Monte Carlo method to jointly determine the best set of archetypes and the number of elements to build the dictionary. Experiments are conducted using synthetic data, a multi-spectral Landsat 7 image of a study area in the Ukraine and the Zurich benchmark data set comprising 20 multispectral Quickbird images. Our results confirm that the proposed approach can learn discriminative features for classification and show better classification results compared to self-taught learning with the original feature representation and compared to randomly initialized archetypal dictionaries.

    @InProceedings{Roscher2016Discriminative,
    Title = {Discriminative Archetypal Self-taught Learning for Multispectral Landcover Classification},
    Author = {Roscher, R. and Wenzel, S. and Waske, B.},
    Booktitle = {Proc. of Pattern Recogniton in Remote Sensing 2016 (PRRS), Workshop at ICPR; to appear in IEEE Xplore},
    Year = {2016},
    Abstract = {Self-taught learning (STL) has become a promising paradigm to exploit unlabeled data for classification. The most commonly used approach to self-taught learning is sparse representation, in which it is assumed that each sample can be represented by a weighted linear combination of elements of a unlabeled dictionary. This paper proposes discriminative archetypal self-taught learning for the application of landcover classification, in which unlabeled discriminative archetypal samples are selected to build a powerful dictionary. Our main contribution is to present an approach which utilizes reversible jump Markov chain Monte Carlo method to jointly determine the best set of archetypes and the number of elements to build the dictionary. Experiments are conducted using synthetic data, a multi-spectral Landsat 7 image of a study area in the Ukraine and the Zurich benchmark data set comprising 20 multispectral Quickbird images. Our results confirm that the proposed approach can learn discriminative features for classification and show better classification results compared to self-taught learning with the original feature representation and compared to randomly initialized archetypal dictionaries.},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2016Discriminative.pdf}
    }

  • T. Schubert, S. Wenzel, R. Roscher, and C. Stachniss, “Investigation of Latent Traces Using Infrared Reflectance Hyperspectral Imaging,” in ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences , 2016, pp. 97-102. doi:10.5194/isprs-annals-III-7-97-2016
    [BibTeX] [PDF]
    The detection of traces is a main task of forensic science. A potential method is hyperspectral imaging (HSI) from which we expect to capture more fluorescence effects than with common Forensic Light Sources (FLS). Specimen of blood, semen and saliva traces in several dilution steps are prepared on cardboard substrate. As our key result we successfully make latent traces visible up to highest available dilution (1:8000). We can attribute most of the detectability to interference of electromagnetic light with the water content of the traces in the Shortwave Infrared region of the spectrum. In a classification task we use several dimensionality reduction methods (PCA and LDA) in combination with a Maximum Likelihood (ML) classifier assuming normally distributed data. Random Forest builds a competitive approach. The classifiers retrieve the exact positions of labeled trace preparation up to highest dilution and determine posterior probabilities. By modeling the classification with a Markov Random Field we obtain smoothed results.

    @InProceedings{Schubert2016Investigation,
    Title = {{Investigation of Latent Traces Using Infrared Reflectance Hyperspectral Imaging}},
    Author = {Schubert, Till and Wenzel, Susanne and Roscher, Ribana and Stachniss, Cyrill},
    Booktitle = {ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences},
    Year = {2016},
    Pages = {97--102},
    Volume = {III-7},
    Abstract = {The detection of traces is a main task of forensic science. A potential method is hyperspectral imaging (HSI) from which we expect to capture more fluorescence effects than with common Forensic Light Sources (FLS). Specimen of blood, semen and saliva traces in several dilution steps are prepared on cardboard substrate. As our key result we successfully make latent traces visible up to highest available dilution (1:8000). We can attribute most of the detectability to interference of electromagnetic light with the water content of the traces in the Shortwave Infrared region of the spectrum. In a classification task we use several dimensionality reduction methods (PCA and LDA) in combination with a Maximum Likelihood (ML) classifier assuming normally distributed data. Random Forest builds a competitive approach. The classifiers retrieve the exact positions of labeled trace preparation up to highest dilution and determine posterior probabilities. By modeling the classification with a Markov Random Field we obtain smoothed results.},
    Doi = {10.5194/isprs-annals-III-7-97-2016},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Schubert2016Investigation.pdf}
    }

2015

  • R. Roscher, C. Römer, B. Waske, and L. Plümer, “Landcover classification with self-taught learning on archetypal dictionaries,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , 2015, pp. 2358-2361. doi:10.1109/IGARSS.2015.7326282
    [BibTeX]
    @InProceedings{Roscher2015Selftaught,
    Title = {Landcover classification with self-taught learning on archetypal dictionaries},
    Author = {Roscher, R. and R\"omer, C. and Waske, B. and Pl\"umer, L.},
    Booktitle = {{IEEE} International Geoscience and Remote Sensing Symposium (IGARSS)},
    Year = {2015},
    Month = {July},
    Pages = {2358-2361},
    Doi = {10.1109/IGARSS.2015.7326282}
    }

  • R. Roscher, B. Uebbing, and J. Kusche, “Spatio-temporal altimeter waveform retracking via sparse representation and conditional random fields,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , 2015, pp. 1234-1237. doi:10.1109/IGARSS.2015.7325996
    [BibTeX]
    @InProceedings{Roscher2015Altimeter,
    Title = {Spatio-temporal altimeter waveform retracking via sparse representation and conditional random fields},
    Author = {Roscher, R. and Uebbing, B. and Kusche, J.},
    Booktitle = {{IEEE} International Geoscience and Remote Sensing Symposium (IGARSS)},
    Year = {2015},
    Month = {July},
    Pages = {1234-1237},
    Doi = {10.1109/IGARSS.2015.7325996}
    }

  • R. Roscher and B. Waske, “Shapelet-Based Sparse Representation for Landcover Classification of Hyperspectral Images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, iss. 3, pp. 1623-1634, 2015. doi:10.1109/TGRS.2015.2484619
    [BibTeX]
    This paper presents a sparse-representation-based classification approach with a novel dictionary construction procedure. By using the constructed dictionary, sophisticated prior knowledge about the spatial nature of the image can be integrated. The approach is based on the assumption that each image patch can be factorized into characteristic spatial patterns, also called shapelets, and patch-specific spectral information. A set of shapelets is learned in an unsupervised way, and spectral information is embodied by training samples. A combination of shapelets and spectral information is represented in an undercomplete spatial-spectral dictionary for each individual patch, where the elements of the dictionary are linearly combined to a sparse representation of the patch. The patch-based classification is obtained by means of the representation error. Experiments are conducted on three well-known hyperspectral image data sets. They illustrate that our proposed approach shows superior results in comparison to sparse-representation-based classifiers that use only limited spatial information and behaves competitively with or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse-representation-based classifiers.

    @Article{Roscher2015Shapelet,
    Title = {Shapelet-Based Sparse Representation for Landcover Classification of Hyperspectral Images},
    Author = {Roscher, R. and Waske, B.},
    Journal = {IEEE Transactions on Geoscience and Remote Sensing},
    Year = {2015},
    Number = {3},
    Pages = {1623--1634},
    Volume = {54},
    Abstract = {This paper presents a sparse-representation-based classification approach with a novel dictionary construction procedure. By using the constructed dictionary, sophisticated prior knowledge about the spatial nature of the image can be integrated. The approach is based on the assumption that each image patch can be factorized into characteristic spatial patterns, also called shapelets, and patch-specific spectral information. A set of shapelets is learned in an unsupervised way, and spectral information is embodied by training samples. A combination of shapelets and spectral information is represented in an undercomplete spatial-spectral dictionary for each individual patch, where the elements of the dictionary are linearly combined to a sparse representation of the patch. The patch-based classification is obtained by means of the representation error. Experiments are conducted on three well-known hyperspectral image data sets. They illustrate that our proposed approach shows superior results in comparison to sparse-representation-based classifiers that use only limited spatial information and behaves competitively with or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse-representation-based classifiers.},
    Doi = {10.1109/TGRS.2015.2484619},
    ISSN = {0196-2892}
    }

2014

  • R. Hagensieker, R. Roscher, and B. Waske, “Texture-based classification of a tropical forest area using multi-temporal ASAR data,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , 2014.
    [BibTeX]
    [none]
    @InProceedings{Hagensieker2014Texture,
    Title = {Texture-based classification of a tropical forest area using multi-temporal ASAR data},
    Author = {Hagensieker, Ron and Roscher, Ribana and Waske, Bj{\"o}rn},
    Booktitle = {IEEE International Geoscience and Remote Sensing Symposium (IGARSS)},
    Year = {2014},
    Abstract = {[none]},
    Owner = {ribana},
    Timestamp = {2014.11.04}
    }

  • K. Herzog, R. Roscher, M. Wieland, A. Kicherer, T. Läbe, W. Förstner, H. Kuhlmann, and R. Töpfer, “Initial steps for high-throughput phenotyping in vineyards,” VITIS – Journal of Grapevine Research, vol. 53, iss. 1, pp. 1-8, 2014.
    [BibTeX]
    The evaluation of phenotypic characters of grape- vines is required directly in the vineyard and is strongly limited by time, costs and the subjectivity of person in charge. Sensor-based techniques are prerequisite to al- low non-invasive phenotyping of individual plant traits, to increase the quantity of object records and to reduce error variation. Thus, a Prototype-Image-Acquisition- System (PIAS) was developed for semi-automated cap- ture of geo-referenced RGB images in an experimental vineyard. Different strategies were tested for image in- terpretation using Matlab. The interpretation of imag- es from the vineyard with the real background is more practice-oriented but requires the calculation of depth maps. Images were utilised to verify the phenotyping results of two semi-automated and one automated pro- totype image interpretation framework. The semi-auto- mated procedures enable contactless and non-invasive detection of bud burst and quantification of shoots at an early developmental stage (BBCH 10) and enable fast and accurate determination of the grapevine berry size at BBCH 89. Depending on the time of image ac- quisition at BBCH 10 up to 94 \% of green shoots were visible in images. The mean berry size (BBCH 89) was recorded non-invasively with a precision of 1 mm.

    @Article{Herzog2014Initial,
    Title = {Initial steps for high-throughput phenotyping in vineyards},
    Author = {Herzog, Katja and Roscher, Ribana and Wieland, Markus and Kicherer,Anna and L\"abe, Thomas and F\"orstner, Wolfgang and Kuhlmann, Heiner and T\"opfer, Reinhard},
    Journal = {VITIS - Journal of Grapevine Research},
    Year = {2014},
    Month = jan,
    Number = {1},
    Pages = {1--8},
    Volume = {53},
    Abstract = {The evaluation of phenotypic characters of grape- vines is required directly in the vineyard and is strongly limited by time, costs and the subjectivity of person in charge. Sensor-based techniques are prerequisite to al- low non-invasive phenotyping of individual plant traits, to increase the quantity of object records and to reduce error variation. Thus, a Prototype-Image-Acquisition- System (PIAS) was developed for semi-automated cap- ture of geo-referenced RGB images in an experimental vineyard. Different strategies were tested for image in- terpretation using Matlab. The interpretation of imag- es from the vineyard with the real background is more practice-oriented but requires the calculation of depth maps. Images were utilised to verify the phenotyping results of two semi-automated and one automated pro- totype image interpretation framework. The semi-auto- mated procedures enable contactless and non-invasive detection of bud burst and quantification of shoots at an early developmental stage (BBCH 10) and enable fast and accurate determination of the grapevine berry size at BBCH 89. Depending on the time of image ac- quisition at BBCH 10 up to 94 \% of green shoots were visible in images. The mean berry size (BBCH 89) was recorded non-invasively with a precision of 1 mm.}
    }

  • A. Kicherer, R. Roscher, K. Herzog, W. Förstner, and R. Töpfer, “Image based Evaluation for the Detection of Cluster Parameters in Grapevine,” in Acta horticulturae , 2014.
    [BibTeX]
    @InProceedings{Kicherer2014Evaluation,
    Title = {Image based Evaluation for the Detection of Cluster Parameters in Grapevine},
    Author = {Kicherer, A. and Roscher, R. and Herzog, K. and F\"orstner, W. and T\"opfer, R.},
    Booktitle = {Acta horticulturae},
    Year = {2014},
    Owner = {ribana},
    Timestamp = {2016.06.20}
    }

  • B. Mack, R. Roscher, and B. Waske, “Can I trust my one-class classification?,” Remote Sensing, vol. 6, iss. 9, pp. 8779-8802, 2014.
    [BibTeX] [PDF]
    Contrary to binary and multi-class classifiers, the purpose of a one-class classifier for remote sensing applications is to map only one specific land use/land cover class of interest. Training these classifiers exclusively requires reference data for the class of interest, while training data for other classes is not required. Thus, the acquisition of reference data can be significantly reduced. However, one-class classification is fraught with uncertainty and full automatization is difficult, due to the limited reference information that is available for classifier training. Thus, a user-oriented one-class classification strategy is proposed, which is based among others on the visualization and interpretation of the one-class classifier outcomes during the data processing. Careful interpretation of the diagnostic plots fosters the understanding of the classification outcome, e.g., the class separability and suitability of a particular threshold. In the absence of complete and representative validation data, which is the fact in the context of a real one-class classification application, such information is valuable for evaluation and improving the classification. The potential of the proposed strategy is demonstrated by classifying different crop types with hyperspectral data from Hyperion.

    @Article{Mack2014Can,
    Title = {Can I trust my one-class classification?},
    Author = {Mack, Benjamin and Roscher, Ribana and Waske, Bj{\"o}rn},
    Journal = {Remote Sensing},
    Year = {2014},
    Number = {9},
    Pages = {8779--8802},
    Volume = {6},
    Abstract = {Contrary to binary and multi-class classifiers, the purpose of a one-class classifier for remote sensing applications is to map only one specific land use/land cover class of interest. Training these classifiers exclusively requires reference data for the class of interest, while training data for other classes is not required. Thus, the acquisition of reference data can be significantly reduced. However, one-class classification is fraught with uncertainty and full automatization is difficult, due to the limited reference information that is available for classifier training. Thus, a user-oriented one-class classification strategy is proposed, which is based among others on the visualization and interpretation of the one-class classifier outcomes during the data processing. Careful interpretation of the diagnostic plots fosters the understanding of the classification outcome, e.g., the class separability and suitability of a particular threshold. In the absence of complete and representative validation data, which is the fact in the context of a real one-class classification application, such information is valuable for evaluation and improving the classification. The potential of the proposed strategy is demonstrated by classifying different crop types with hyperspectral data from Hyperion.},
    Owner = {ribana},
    Timestamp = {2014.11.04},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Mack2014Can.pdf}
    }

  • R. Roscher, K. Herzog, A. Kunkel, A. Kicherer, R. Töpfer, and W. Förstner, “Automated image analysis framework for high-throughput determination of grapevine berry sizes using conditional random fields,” Computers and Electronics in Agriculture, vol. 100, pp. 148-158, 2014. doi:10.1016/j.compag.2013.11.008
    [BibTeX]
    @Article{Roscher2014Automated,
    Title = {Automated image analysis framework for high-throughput determination of grapevine berry sizes using conditional random fields},
    Author = {Roscher, Ribana and Herzog, Katja and Kunkel, Annemarie and Kicherer, Anna and T{\"o}pfer, Reinhard and F{\"o}rstner, Wolfgang},
    Journal = {Computers and Electronics in Agriculture},
    Year = {2014},
    Pages = {148--158},
    Volume = {100},
    Doi = {10.1016/j.compag.2013.11.008},
    Publisher = {Elsevier}
    }

  • R. Roscher and B. Waske, “Shapelet-based sparse image representation for landcover classification of hyperspectral data,” in IAPR Workshop on Pattern Recognition in Remote Sensing , 2014, pp. 1-6.
    [BibTeX] [PDF]
    This paper presents a novel sparse representation-based classifier for landcover mapping of hyperspectral image data. Each image patch is factorized into segmentation patterns, also called shapelets, and patch-specific spectral features. The combination of both is represented in a patch-specific spatial-spectral dictionary, which is used for a sparse coding procedure for the reconstruction and classification of image patches. Hereby, each image patch is sparsely represented by a linear combination of elements out of the dictionary. The set of shapelets is specifically learned for each image in an unsupervised way in order to capture the image structure. The spectral features are assumed to be the training data. The experiments show that the proposed approach shows superior results in comparison to sparse-representation based classifiers that use no or only limited spatial information and behaves competitive or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse representation-based classifiers.

    @InProceedings{Roscher2014Shapelet,
    Title = {Shapelet-based sparse image representation for landcover classification of hyperspectral data},
    Author = {Roscher, Ribana and Waske, Bj{\"o}rn},
    Booktitle = {IAPR Workshop on Pattern Recognition in Remote Sensing},
    Year = {2014},
    Pages = {1--6},
    Abstract = {This paper presents a novel sparse representation-based classifier for landcover mapping of hyperspectral image data. Each image patch is factorized into segmentation patterns, also called shapelets, and patch-specific spectral features. The combination of both is represented in a patch-specific spatial-spectral dictionary, which is used for a sparse coding procedure for the reconstruction and classification of image patches. Hereby, each image patch is sparsely represented by a linear combination of elements out of the dictionary. The set of shapelets is specifically learned for each image in an unsupervised way in order to capture the image structure. The spectral features are assumed to be the training data. The experiments show that the proposed approach shows superior results in comparison to sparse-representation based classifiers that use no or only limited spatial information and behaves competitive or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse representation-based classifiers.},
    Owner = {ribana},
    Timestamp = {2014.11.04},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2014Shapelet.pdf}
    }

  • R. Roscher and B. Waske, “Superpixel-based classification of hyperspectral data using sparse representation and conditional random fields,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , 2014.
    [BibTeX] [PDF]
    This paper presents a superpixel-based classifier for landcover mapping of hyperspectral image data. The approach relies on the sparse representation of each pixel by a weighted linear combination of the training data. Spatial information is incorporated by using a coarse patch-based neighborhood around each pixel as well as data-adapted superpixels. The classification is done via a hierarchical conditional random field, which utilizes the sparse-representation output and models spatial and hierarchical structures in the hyperspectral image. The experiments show that the proposed approach results in superior accuracies in comparison to sparse-representation based classifiers that solely use a patch-based neighborhood.

    @InProceedings{Roscher2014Superpixel,
    Title = {Superpixel-based classification of hyperspectral data using sparse representation and conditional random fields},
    Author = {Roscher, Ribana and Waske, Bj{\"o}rn},
    Booktitle = {{IEEE} International Geoscience and Remote Sensing Symposium (IGARSS)},
    Year = {2014},
    Abstract = {This paper presents a superpixel-based classifier for landcover mapping of hyperspectral image data. The approach relies on the sparse representation of each pixel by a weighted linear combination of the training data. Spatial information is incorporated by using a coarse patch-based neighborhood around each pixel as well as data-adapted superpixels. The classification is done via a hierarchical conditional random field, which utilizes the sparse-representation output and models spatial and hierarchical structures in the hyperspectral image. The experiments show that the proposed approach results in superior accuracies in comparison to sparse-representation based classifiers that solely use a patch-based neighborhood.},
    Owner = {ribana},
    Timestamp = {2014.11.04},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2014Superpixel.pdf}
    }

2013

  • A. Kicherer, R. Roscher, K. Herzog, S. Šimon, W. Förstner, and R. Töpfer, “BAT (Berry Analysis Tool): A high-throughput image interpretation tool to acquire the number, diameter, and volume of grapevine berries,” Vitis, vol. 52, iss. 3, pp. 129-135, 2013.
    [BibTeX]
    QTL-analysis (quantitative trait loci) and marker development rely on efficient phenotyping techniques. Objectivity and precision of a phenotypic data evaluation is crucial but time consuming. In the present study a high-throughput image interpretation tool was developed to acquire automatically number, size, and volume of grape berries from RGB (red-green-blue) images. Individual berries of one cluster were placed on a defined construction to take a RGB image from the top. The image interpretation of one dataset with an arbitrary number of images occurs automatically by starting the BAT (Berry-Analysis-Tool) developed in MATLAB. For validation of results, the number of berries was counted and their size was measured using a digital calliper. A measuring cylinder was used to determine reliably the berry volume by displacement of water. All placed berries could be counted by BAT 100\A0\% correctly. Manual ratings compared with BAT ratings showed strong correlation of r\A0=\A00,964 for mean berry diameter/image and r\A0=\A00.984 for berry volume.

    @Article{Kicherer2013,
    Title = {BAT (Berry Analysis Tool): A high-throughput image interpretation tool to acquire the number, diameter, and volume of grapevine berries},
    Author = {Kicherer, A. and Roscher, R. and Herzog, K. and {\vS}imon, S. and F\"orstner, W. and T\"opfer, R.},
    Journal = {Vitis},
    Year = {2013},
    Number = {3},
    Pages = {129-135},
    Volume = {52},
    Abstract = {QTL-analysis (quantitative trait loci) and marker development rely on efficient phenotyping techniques. Objectivity and precision of a phenotypic data evaluation is crucial but time consuming. In the present study a high-throughput image interpretation tool was developed to acquire automatically number, size, and volume of grape berries from RGB (red-green-blue) images. Individual berries of one cluster were placed on a defined construction to take a RGB image from the top. The image interpretation of one dataset with an arbitrary number of images occurs automatically by starting the BAT (Berry-Analysis-Tool) developed in MATLAB. For validation of results, the number of berries was counted and their size was measured using a digital calliper. A measuring cylinder was used to determine reliably the berry volume by displacement of water. All placed berries could be counted by BAT 100\A0\% correctly. Manual ratings compared with BAT ratings showed strong correlation of r\A0=\A00,964 for mean berry diameter/image and r\A0=\A00.984 for berry volume.},
    Owner = {ribana1},
    Timestamp = {2013.08.14}
    }

2012

  • R. Roscher, “Sequential Learning using Incremental Import Vector Machines for Semantic Segmentation,” PhD Thesis, 2012.
    [BibTeX] [PDF]
    We propose an innovative machine learning algorithm called incremental import vector machines that is used for classification purposes. The classifier is specifically designed for the task of sequential learning, in which the data samples are successively presented to the classifier. The motivation for our work comes from the effort to formulate a classifier that can manage the major challenges of sequential learning problems, while being a powerful classifier in terms of classification accuracy, efficiency and meaningful output. One challenge of sequential learning is that data samples are not completely available to the learner at a given point of time and generally, waiting for a representative number of data is undesirable and impractical. Thus, in order to allow for a classification of given data samples at any time, the learning phase of the classifier model needs to start immediately, even if not all training samples are available. Another challenge is that the number of sequential arriving data samples can be very large or even infinite and thus, not all samples can be stored. Furthermore, the distribution of the sample can vary over time and the classifier model needs to remain stable and unchanged to irrelevant samples while being plastic to new, important samples. Therefore our key contribution is to develop, analyze and evaluate a powerful incremental learner for sequential learning which we call incremental import vector machines (I2VMs). The classifier is based on the batch machine learning algorithm import vector machines, which was developed by Zhu and Hastie (2005). I2VM is a kernel-based, discriminative classifier and thus, is able to deal with complex data distributions. Additionally ,the learner is sparse for an efficient training and testing and has a probabilistic output. A key achievement of this thesis is the verification and analysis of the discriminative and reconstructive model components of IVM and I2VM. While discriminative classifiers try to separate the classes as well as possible, classifiers with a reconstructive component aspire to have a high information content in order to approximate the distribution of the data samples. Both properties are necessary for a powerful incremental classifier. A further key achievement is the formulation of the incremental learning strategy of I2VM. The strategy deals with adding and removing data samples and the update of the current set of model parameters. Furthermore, also new classes and features can be incorporated. The learning strategy adapts the model continuously, while keeping it stable and efficient. In our experiments we use I2VM for the semantic segmentation of images from an image database, for large area land cover classification of overlapping remote sensing images and for object tracking in image sequences. We show that I2VM results in superior or competitive classification accuracies to comparable classifiers. A substantial achievement of the thesis is that I2VM’s performance is independent of the ordering of the data samples and a reconsidering of already encountered samples for learning is not necessary. A further achievement is that I2VM is able to deal with very long data streams without a loss in the efficiency. Furthermore, as another achievement, we show that I2VM provide reliable posterior probabilities since samples with high class probabilities are accurately classified, whereas relatively low class probabilities are more likely referred to misclassified samples.

    @PhdThesis{Roscher2012Sequential,
    Title = {Sequential Learning using Incremental Import Vector Machines for Semantic Segmentation},
    Author = {Roscher, Ribana},
    School = {Department of Photogrammetry, University of Bonn},
    Year = {2012},
    Abstract = {We propose an innovative machine learning algorithm called incremental import vector machines that is used for classification purposes. The classifier is specifically designed for the task of sequential learning, in which the data samples are successively presented to the classifier. The motivation for our work comes from the effort to formulate a classifier that can manage the major challenges of sequential learning problems, while being a powerful classifier in terms of classification accuracy, efficiency and meaningful output. One challenge of sequential learning is that data samples are not completely available to the learner at a given point of time and generally, waiting for a representative number of data is undesirable and impractical. Thus, in order to allow for a classification of given data samples at any time, the learning phase of the classifier model needs to start immediately, even if not all training samples are available. Another challenge is that the number of sequential arriving data samples can be very large or even infinite and thus, not all samples can be stored. Furthermore, the distribution of the sample can vary over time and the classifier model needs to remain stable and unchanged to irrelevant samples while being plastic to new, important samples. Therefore our key contribution is to develop, analyze and evaluate a powerful incremental learner for sequential learning which we call incremental import vector machines (I2VMs). The classifier is based on the batch machine learning algorithm import vector machines, which was developed by Zhu and Hastie (2005). I2VM is a kernel-based, discriminative classifier and thus, is able to deal with complex data distributions. Additionally ,the learner is sparse for an efficient training and testing and has a probabilistic output. A key achievement of this thesis is the verification and analysis of the discriminative and reconstructive model components of IVM and I2VM. While discriminative classifiers try to separate the classes as well as possible, classifiers with a reconstructive component aspire to have a high information content in order to approximate the distribution of the data samples. Both properties are necessary for a powerful incremental classifier. A further key achievement is the formulation of the incremental learning strategy of I2VM. The strategy deals with adding and removing data samples and the update of the current set of model parameters. Furthermore, also new classes and features can be incorporated. The learning strategy adapts the model continuously, while keeping it stable and efficient. In our experiments we use I2VM for the semantic segmentation of images from an image database, for large area land cover classification of overlapping remote sensing images and for object tracking in image sequences. We show that I2VM results in superior or competitive classification accuracies to comparable classifiers. A substantial achievement of the thesis is that I2VM's performance is independent of the ordering of the data samples and a reconsidering of already encountered samples for learning is not necessary. A further achievement is that I2VM is able to deal with very long data streams without a loss in the efficiency. Furthermore, as another achievement, we show that I2VM provide reliable posterior probabilities since samples with high class probabilities are accurately classified, whereas relatively low class probabilities are more likely referred to misclassified samples.},
    City = {Bonn},
    Url = {http://hss.ulb.uni-bonn.de/2012/3009/3009.htm}
    }

  • R. Roscher, W. Förstner, and B. Waske, “I²VM: Incremental import vector machines,” Image and Vision Computing, vol. 30, iss. 4-5, pp. 263-278, 2012. doi:10.1016/j.imavis.2012.04.004
    [BibTeX]
    We introduce an innovative incremental learner called incremental import vector machines ((IVM)-V-2). The kernel-based discriminative approach is able to deal with complex data distributions. Additionally, the learner is sparse for an efficient training and testing and has a probabilistic output. We particularly investigate the reconstructive component of import vector machines, in order to use it for robust incremental teaming. By performing incremental update steps, we are able to add and remove data samples, as well as update the current set of model parameters for incremental learning. By using various standard benchmarks, we demonstrate how (IVM)-V-2 is competitive or superior to other incremental methods. It is also shown that our approach is capable of managing concept-drifts in the data distributions. (C) 2012 Elsevier B.V. All rights reserved.

    @Article{Roscher2012I2VM,
    Title = {I²VM: Incremental import vector machines},
    Author = {Roscher, Ribana and F\"orstner, Wolfgang and Waske, Bj\"orn},
    Journal = {Image and Vision Computing},
    Year = {2012},
    Month = may,
    Number = {4-5},
    Pages = {263--278},
    Volume = {30},
    Abstract = {We introduce an innovative incremental learner called incremental import vector machines ((IVM)-V-2). The kernel-based discriminative approach is able to deal with complex data distributions. Additionally, the learner is sparse for an efficient training and testing and has a probabilistic output. We particularly investigate the reconstructive component of import vector machines, in order to use it for robust incremental teaming. By performing incremental update steps, we are able to add and remove data samples, as well as update the current set of model parameters for incremental learning. By using various standard benchmarks, we demonstrate how (IVM)-V-2 is competitive or superior to other incremental methods. It is also shown that our approach is capable of managing concept-drifts in the data distributions. (C) 2012 Elsevier B.V. All rights reserved.},
    Doi = {10.1016/j.imavis.2012.04.004},
    Owner = {waske},
    Sn = {0262-8856},
    Tc = {0},
    Timestamp = {2012.09.04},
    Ut = {WOS:000305726700001},
    Z8 = {0},
    Z9 = {0},
    Zb = {0}
    }

  • R. Roscher, J. Siegemund, F. Schindler, and W. Förstner, “Object Tracking by Segmentation Using Incremental Import Vector Machines,” Department of Photogrammetry, University of Bonn 2012.
    [BibTeX] [PDF]
    We propose a framework for object tracking in image sequences, following the concept of tracking-by-segmentation. The separation of object and background is achieved by a consecutive semantic superpixel segmentation of the images, yielding tight object boundaries. I.e., in the first image a model of the object’s characteristics is learned from an initial, incomplete annotation. This model is used to classify the superpixels of subsequent images to object and background employing graph-cut. We assume the object boundaries to be tight-fitting and the object motion within the image to be affine. To adapt the model to radiometric and geometric changes we utilize an incremental learner in a co-training scheme. We evaluate our tracking framework qualitatively and quantitatively on several image sequences.

    @TechReport{Roscher2012Object,
    Title = {Object Tracking by Segmentation Using Incremental Import Vector Machines},
    Author = {Roscher, Ribana and Siegemund, Jan and Schindler, Falko and F\"orstner, Wolfgang},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2012},
    Abstract = {We propose a framework for object tracking in image sequences, following the concept of tracking-by-segmentation. The separation of object and background is achieved by a consecutive semantic superpixel segmentation of the images, yielding tight object boundaries. I.e., in the first image a model of the object's characteristics is learned from an initial, incomplete annotation. This model is used to classify the superpixels of subsequent images to object and background employing graph-cut. We assume the object boundaries to be tight-fitting and the object motion within the image to be affine. To adapt the model to radiometric and geometric changes we utilize an incremental learner in a co-training scheme. We evaluate our tracking framework qualitatively and quantitatively on several image sequences.},
    City = {Bonn},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2012Object.pdf}
    }

  • R. Roscher, B. Waske, and W. Förstner, “Evaluation of Import Vector Machines for Classifying Hyperspectral Data,” Department of Photogrammetry, University of Bonn 2012.
    [BibTeX] [PDF]
    We evaluate the performance of Import Vector Machines (IVM),a sparse Kernel Logistic Regression approach, for the classification of hyperspectral data. The IVM classifier is applied on two different data sets, using different number of training samples. The performance of IVM to Support Vector Machines (SVM) is compared in terms of accuracy and sparsity. Moreover, the impact of the training sample set on the accuracy and stability of IVM was investigated. The results underline that the IVM perform similar when compared to the popular SVM in terms of accuracy. Moreover, the number of import vectors from the IVM is significantly lower when compared to the number of support vectors from the SVM. Thus, the classification process of the IVM is faster. These findings are independent from the study site, the number of training samples and specific classes. Consequently, the proposed IVM approach is a promising classification method for hyperspectral imagery.

    @TechReport{Roscher2012Evaluation,
    Title = {Evaluation of Import Vector Machines for Classifying Hyperspectral Data},
    Author = {Roscher, Ribana and Waske, Bj\"orn and F\"orstner, Wolfgang},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2012},
    Abstract = {We evaluate the performance of Import Vector Machines (IVM),a sparse Kernel Logistic Regression approach, for the classification of hyperspectral data. The IVM classifier is applied on two different data sets, using different number of training samples. The performance of IVM to Support Vector Machines (SVM) is compared in terms of accuracy and sparsity. Moreover, the impact of the training sample set on the accuracy and stability of IVM was investigated. The results underline that the IVM perform similar when compared to the popular SVM in terms of accuracy. Moreover, the number of import vectors from the IVM is significantly lower when compared to the number of support vectors from the SVM. Thus, the classification process of the IVM is faster. These findings are independent from the study site, the number of training samples and specific classes. Consequently, the proposed IVM approach is a promising classification method for hyperspectral imagery.},
    City = {Bonn},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2012Evaluation.pdf}
    }

  • R. Roscher, B. Waske, and W. Förstner, “Incremental Import Vector Machines for Classifying Hyperspectral Data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, iss. 9, pp. 3463-3473, 2012. doi:10.1109/TGRS.2012.2184292
    [BibTeX]
    In this paper, we propose an incremental learning strategy for import vector machines (IVM), which is a sparse kernel logistic regression approach. We use the procedure for the concept of self-training for sequential classification of hyperspectral data. The strategy comprises the inclusion of new training samples to increase the classification accuracy and the deletion of noninformative samples to be memory and runtime efficient. Moreover, we update the parameters in the incremental IVM model without retraining from scratch. Therefore, the incremental classifier is able to deal with large data sets. The performance of the IVM in comparison to support vector machines (SVM) is evaluated in terms of accuracy, and experiments are conducted to assess the potential of the probabilistic outputs of the IVM. Experimental results demonstrate that the IVM and SVM perform similar in terms of classification accuracy. However, the number of import vectors is significantly lower when compared to the number of support vectors, and thus, the computation time during classification can be decreased. Moreover, the probabilities provided by IVM are more reliable, when compared to the probabilistic information, derived from an SVM’s output. In addition, the proposed self-training strategy can increase the classification accuracy. Overall, the IVM and its incremental version is worthwhile for the classification of hyperspectral data.

    @Article{Roscher2012Incremental,
    Title = {Incremental Import Vector Machines for Classifying Hyperspectral Data},
    Author = {Roscher, Ribana and Waske, Bj\"orn and F\"orstner, Wolfgang},
    Journal = {IEEE Transactions on Geoscience and Remote Sensing},
    Year = {2012},
    Month = sep,
    Number = {9},
    Pages = {3463--3473},
    Volume = {50},
    Abstract = {In this paper, we propose an incremental learning strategy for import vector machines (IVM), which is a sparse kernel logistic regression approach. We use the procedure for the concept of self-training for sequential classification of hyperspectral data. The strategy comprises the inclusion of new training samples to increase the classification accuracy and the deletion of noninformative samples to be memory and runtime efficient. Moreover, we update the parameters in the incremental IVM model without retraining from scratch. Therefore, the incremental classifier is able to deal with large data sets. The performance of the IVM in comparison to support vector machines (SVM) is evaluated in terms of accuracy, and experiments are conducted to assess the potential of the probabilistic outputs of the IVM. Experimental results demonstrate that the IVM and SVM perform similar in terms of classification accuracy. However, the number of import vectors is significantly lower when compared to the number of support vectors, and thus, the computation time during classification can be decreased. Moreover, the probabilities provided by IVM are more reliable, when compared to the probabilistic information, derived from an SVM's output. In addition, the proposed self-training strategy can increase the classification accuracy. Overall, the IVM and its incremental version is worthwhile for the classification of hyperspectral data.},
    Doi = {10.1109/TGRS.2012.2184292},
    ISSN = {0196-2892},
    Owner = {waske},
    Timestamp = {2012.09.05}
    }

2011

  • M. Hans and R. Roscher, “Zuordnen radiometrischer Informationen zu Laserscandaten von Weintrauben,” Department of Photogrammetry, University of Bonn 2011.
    [BibTeX] [PDF]
    In diesem Report stellen wir zwei Verfahren vor, die radiometrische Informationen 3D-Scandaten zuordnen. Radiometrische Informationen unterstützen und verbessern die Anwendungen der Merkmalserfassung von Objekten, da sie weitere Kenntnisse über das gescannte Objekt liefern.

    @TechReport{Hans2011Zuordnen,
    Title = {Zuordnen radiometrischer Informationen zu Laserscandaten von Weintrauben},
    Author = {Hans, Mathias and Roscher, Ribana},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2011},
    Abstract = {In diesem Report stellen wir zwei Verfahren vor, die radiometrische Informationen 3D-Scandaten zuordnen. Radiometrische Informationen unterst\"utzen und verbessern die Anwendungen der Merkmalserfassung von Objekten, da sie weitere Kenntnisse \"uber das gescannte Objekt liefern.},
    City = {Bonn},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Hans2011Zuordnen.pdf}
    }

  • R. Roscher, F. Schindler, and W. Förstner, “What would you look like in Springfield? Linear Transformations between High-Dimensional Spaces,” Department of Photogrammetry, University of Bonn 2011.
    [BibTeX] [PDF]
    High-dimensional data structures occur in many fields of computer vision and machine learning. Transformation between two high-dimensional spaces usually involves the determination of a large amount of parameters and requires much labeled data to be given. There is much interest in reducing dimensionality if a lower-dimensional structure is underlying the data points. We present a procedure to enable the determination of a low-dimensional, projective transformation between two data sets, making use of state-of-the-art dimensional reduction algorithms. We evaluate multiple algorithms during several experiments with different objectives. We demonstrate the use of this procedure for applications like classification and assignments between two given data sets. Our procedure is semi-supervised due to the fact that all labeled and unlabeled points are used for the dimensionality reduction, but only few them have to be labeled. Using test data we evaluate the quantitative and qualitative performance of different algorithms with respect to the classification and assignment task. We show that with these algorithms and our transformation approach high-dimensional data sets can be related to each other. Finally we can use this procedure to match real world facial images with cartoon images from Springfield, home town of the famous Simpsons.

    @TechReport{Roscher2011What,
    Title = {What would you look like in Springfield? Linear Transformations between High-Dimensional Spaces},
    Author = {Roscher, Ribana and Schindler, Falko and F\"orstner, Wolfgang},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2011},
    Abstract = {High-dimensional data structures occur in many fields of computer vision and machine learning. Transformation between two high-dimensional spaces usually involves the determination of a large amount of parameters and requires much labeled data to be given. There is much interest in reducing dimensionality if a lower-dimensional structure is underlying the data points. We present a procedure to enable the determination of a low-dimensional, projective transformation between two data sets, making use of state-of-the-art dimensional reduction algorithms. We evaluate multiple algorithms during several experiments with different objectives. We demonstrate the use of this procedure for applications like classification and assignments between two given data sets. Our procedure is semi-supervised due to the fact that all labeled and unlabeled points are used for the dimensionality reduction, but only few them have to be labeled. Using test data we evaluate the quantitative and qualitative performance of different algorithms with respect to the classification and assignment task. We show that with these algorithms and our transformation approach high-dimensional data sets can be related to each other. Finally we can use this procedure to match real world facial images with cartoon images from Springfield, home town of the famous Simpsons.},
    City = {Bonn},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2011What.pdf}
    }

  • R. Roscher, B. Waske, and W. Förstner, “Incremental import vector machines for large area land cover classification,” in IEEE International Conference on Computer Vision Workshops (ICCV Workshops) , 2011. doi:10.1109/ICCVW.2011.6130249
    [BibTeX]
    The classification of large areas consisting of multiple scenes is challenging regarding the handling of large and therefore mostly inhomogeneous data sets. Moreover, large data sets demand for computational efficient methods. We propose a method, which enables the efficient multi-class classification of large neighboring Landsat scenes. We use an incremental realization of the import vector machines, called I2VM, in combination with self-training to update an initial learned classifier with new training data acquired in the overlapping areas between neighboring Landsat scenes. We show in our experiments, that I2VM is a suitable classifier for large area land cover classification.

    @InProceedings{Roscher2011Incremental,
    Title = {Incremental import vector machines for large area land cover classification},
    Author = {Roscher, Ribana and Waske, Bj\"orn and F\"orstner, Wolfgang},
    Booktitle = {{IEEE} International Conference on Computer Vision Workshops (ICCV Workshops)},
    Year = {2011},
    Abstract = {The classification of large areas consisting of multiple scenes is challenging regarding the handling of large and therefore mostly inhomogeneous data sets. Moreover, large data sets demand for computational efficient methods. We propose a method, which enables the efficient multi-class classification of large neighboring Landsat scenes. We use an incremental realization of the import vector machines, called I2VM, in combination with self-training to update an initial learned classifier with new training data acquired in the overlapping areas between neighboring Landsat scenes. We show in our experiments, that I2VM is a suitable classifier for large area land cover classification.},
    Doi = {10.1109/ICCVW.2011.6130249},
    Keywords = {incremental import vector machines;inhomogeneous data sets;land cover classification;neighboring Landsat scenes;scenes classification;training data acquisition;data acquisition;geophysical image processing;image classification;natural scenes;support vector machines;terrain mapping;},
    Owner = {waske},
    Timestamp = {2012.09.05}
    }

  • B. Waske, R. Roscher, and S. Klemenjak, “Import Vector Machines Based Classification of Multisensor Remote Sensing Data,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , 2011. doi:10.1109/IGARSS.2011.6049829
    [BibTeX]
    The classification of multisensor data sets, consisting of multitemporal SAR data and multispectral is addressed. In the present study, Import Vector Machines (IVM) are applied on two data sets, consisting of (i) Envisat ASAR/ERS-2 SAR data and a Landsat 5 TM scene, and (h) TerraSAR-X data and a RapidEye scene. The performance of IVM for classifying multisensor data is evaluated and the method is compared to Support Vector Machines (SVM) in terms of accuracy and complexity. In general, the experimental results demonstrate that the classification accuracy is improved by the multisensor data set. Moreover, IVM and SVM perform similar in terms of the classification accuracy. However, the number of import vectors is considerably less than the number of support vectors, and thus the computation time of the IVM classification is lower. IVM can directly be applied to the multi-class problems and provide probabilistic outputs. Overall IVM constitutes a feasible method and alternative to SVM.

    @InProceedings{Waske2011Import,
    Title = {Import Vector Machines Based Classification of Multisensor Remote Sensing Data},
    Author = {Waske, Bj\"orn and Roscher, Ribana and Klemenjak, Sascha},
    Booktitle = {IEEE International Geoscience and Remote Sensing Symposium (IGARSS)},
    Year = {2011},
    Abstract = {The classification of multisensor data sets, consisting of multitemporal SAR data and multispectral is addressed. In the present study, Import Vector Machines (IVM) are applied on two data sets, consisting of (i) Envisat ASAR/ERS-2 SAR data and a Landsat 5 TM scene, and (h) TerraSAR-X data and a RapidEye scene. The performance of IVM for classifying multisensor data is evaluated and the method is compared to Support Vector Machines (SVM) in terms of accuracy and complexity. In general, the experimental results demonstrate that the classification accuracy is improved by the multisensor data set. Moreover, IVM and SVM perform similar in terms of the classification accuracy. However, the number of import vectors is considerably less than the number of support vectors, and thus the computation time of the IVM classification is lower. IVM can directly be applied to the multi-class problems and provide probabilistic outputs. Overall IVM constitutes a feasible method and alternative to SVM.},
    Doi = {10.1109/IGARSS.2011.6049829},
    Keywords = {Envisat ASAR ERS-2 SAR data;IVM;Landsat 5 TM scene;RapidEye scene;SVM comparison;TerraSAR-X data;computation time;data classification;import vector machines;multisensor remote sensing data;multispectral data;multitemporal SAR data;support vector machines;geophysical image processing;image classification;knowledge engineering;radar imaging;remote sensing by radar;spaceborne radar;synthetic aperture radar;}
    }

2010

  • R. Roscher, F. Schindler, and W. Förstner, “High Dimensional Correspondences from Low Dimensional Manifolds — An Empirical Comparison of Graph-based Dimensionality Reduction Algorithms,” in The 3rd International Workshop on Subspace Methods, in conjunction with ACCV2010 , 2010, p. 10. doi:10.1007/978-3-642-22819-3_34
    [BibTeX] [PDF]
    We discuss the utility of dimensionality reduction algorithms to put data points in high dimensional spaces into correspondence by learning a transformation between assigned data points on a lower dimensional structure. We assume that similar high dimensional feature spaces are characterized by a similar underlying low dimensional structure. To enable the determination of an affine transformation between two data sets we make use of well-known dimensional reduction algorithms. We demonstrate this procedure for applications like classification and assignments between two given data sets and evaluate six well-known algorithms during several experiments with different objectives. We show that with these algorithms and our transformation approach high dimensional data sets can be related to each other. We also show that linear methods turn out to be more suitable for assignment tasks, whereas graph-based methods appear to be superior for classification tasks.

    @InProceedings{Roscher2010High,
    Title = {High Dimensional Correspondences from Low Dimensional Manifolds -- An Empirical Comparison of Graph-based Dimensionality Reduction Algorithms},
    Author = {Roscher, Ribana and Schindler, Falko and F\"orstner, Wolfgang},
    Booktitle = {The 3rd International Workshop on Subspace Methods, in conjunction with ACCV2010},
    Year = {2010},
    Note = {Queenstown, New Zealand},
    Pages = {10},
    Abstract = {We discuss the utility of dimensionality reduction algorithms to put data points in high dimensional spaces into correspondence by learning a transformation between assigned data points on a lower dimensional structure. We assume that similar high dimensional feature spaces are characterized by a similar underlying low dimensional structure. To enable the determination of an affine transformation between two data sets we make use of well-known dimensional reduction algorithms. We demonstrate this procedure for applications like classification and assignments between two given data sets and evaluate six well-known algorithms during several experiments with different objectives. We show that with these algorithms and our transformation approach high dimensional data sets can be related to each other. We also show that linear methods turn out to be more suitable for assignment tasks, whereas graph-based methods appear to be superior for classification tasks.},
    Doi = {10.1007/978-3-642-22819-3_34},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2010High.pdf;Poster:Roscher2010High_Poster.pdf}
    }

  • R. Roscher, B. Waske, and W. Förstner, “Kernel Discriminative Random Fields for land cover classification,” in IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS) , 2010. doi:10.1109/PRRS.2010.5742801
    [BibTeX] [PDF]
    Logistic Regression has become a commonly used classifier, not only due to its probabilistic output and its direct usage in multi-class cases. We use a sparse Kernel Logistic Regression approach – the Import Vector Machines – for land cover classification. We improve our segmentation results applying a Discriminative Random Field framework on the probabilistic classification output. We consider the performance regarding to the classification accuracy and the complexity and compare it to the Gaussian Maximum Likelihood classification and the Support Vector Machines.

    @InProceedings{Roscher2010Kernel,
    Title = {Kernel Discriminative Random Fields for land cover classification},
    Author = {Roscher, Ribana and Waske, Bj\"orn and F\"orstner, Wolfgang},
    Booktitle = {IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)},
    Year = {2010},
    Note = {Istanbul, Turkey},
    Abstract = {Logistic Regression has become a commonly used classifier, not only due to its probabilistic output and its direct usage in multi-class cases. We use a sparse Kernel Logistic Regression approach - the Import Vector Machines - for land cover classification. We improve our segmentation results applying a Discriminative Random Field framework on the probabilistic classification output. We consider the performance regarding to the classification accuracy and the complexity and compare it to the Gaussian Maximum Likelihood classification and the Support Vector Machines.},
    Doi = {10.1109/PRRS.2010.5742801},
    Keywords = {Gaussian maximum likelihood classification;image segmentation;import vector machine;kernel discriminative random fields;land cover classification;logistic regression;probabilistic classification;support vector machines;geophysical image processing;image classification;image segmentation;support vector machines;terrain mapping;},
    Owner = {waske},
    Timestamp = {2012.09.05},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2010Kernel.pdf;Slides:Roscher2010Kernel_Slides.pdf}
    }

2009

  • M. Drauschke, R. Roscher, T. Läbe, and W. Förstner, “Improving Image Segmentation using Multiple View Analysis,” in Object Extraction for 3D City Models, Road Databases and Traffic Monitoring – Concepts, Algorithms and Evaluatin (CMRT09) , 2009, pp. 211-216.
    [BibTeX] [PDF]
    In our contribution, we improve image segmentation by integrating depth information from multi-view analysis. We assume the object surface in each region can be represented by a low order polynomial, and estimate the best fitting parameters of a plane using those points of the point cloud, which are mapped to the specific region. We can merge adjacent image regions, which cannot be distinguished geometrically. We demonstrate the approach for finding spatially planar regions on aerial images. Furthermore, we discuss the possibilities of extending of our approach towards segmenting terrestrial facade images.

    @InProceedings{Drauschke2009Improving,
    Title = {Improving Image Segmentation using Multiple View Analysis},
    Author = {Drauschke, Martin and Roscher, Ribana and L\"abe, Thomas and F\"orstner, Wolfgang},
    Booktitle = {Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms and Evaluatin (CMRT09)},
    Year = {2009},
    Pages = {211-216},
    Abstract = {In our contribution, we improve image segmentation by integrating depth information from multi-view analysis. We assume the object surface in each region can be represented by a low order polynomial, and estimate the best fitting parameters of a plane using those points of the point cloud, which are mapped to the specific region. We can merge adjacent image regions, which cannot be distinguished geometrically. We demonstrate the approach for finding spatially planar regions on aerial images. Furthermore, we discuss the possibilities of extending of our approach towards segmenting terrestrial facade images.},
    City = {Paris},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Drauschke2009Improving.pdf}
    }

  • R. Roscher and W. Förstner, “Multiclass Bounded Logistic Regression — Efficient Regularization with Interior Point Method,” Department of Photogrammetry, University of Bonn, TR-IGG-P-2009-02, 2009.
    [BibTeX] [PDF]
    Logistic regression has been widely used in classi cation tasks for many years. Its optimization in case of linear separable data has received extensive study due to the problem of a monoton likelihood. This paper presents a new approach, called bounded logistic regression (BLR), by solving the logistic regression as a convex optimization problem with constraints. The paper tests the accuracy of BLR by evaluating nine well-known datasets and compares it to the closely related support vector machine approach (SVM).

    @TechReport{Roscher2009Multiclass,
    Title = {Multiclass Bounded Logistic Regression -- Efficient Regularization with Interior Point Method},
    Author = {Roscher, Ribana and F\"orstner, Wolfgang},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2009},
    Number = {TR-IGG-P-2009-02},
    Abstract = {Logistic regression has been widely used in classication tasks for many years. Its optimization in case of linear separable data has received extensive study due to the problem of a monoton likelihood. This paper presents a new approach, called bounded logistic regression (BLR), by solving the logistic regression as a convex optimization problem with constraints. The paper tests the accuracy of BLR by evaluating nine well-known datasets and compares it to the closely related support vector machine approach (SVM).},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2009Multiclass.pdf}
    }

2008

  • R. Roscher, “Bestimmung von 3D-Merkmalen von Bildregionen aus Stereobildern,” Department of Photogrammetry, University of Bonn, TR-IGG-P-2008-05, 2008.
    [BibTeX] [PDF]
    Dieser Report erläutert die Bestimmung von 3D-Merkmalen von Bildregionen durch Zuordnung von Bildpunkten in diesen Regionen zu den Objektpunkten in einer Punktwolke.Die Umsetzung erfolgt in einer grafischen Benutzeroberfläche in Matlab, deren Bedienung in diesem Report veranschaulicht werden soll.

    @TechReport{Roscher2008Bestimmung,
    Title = {Bestimmung von 3D-Merkmalen von Bildregionen aus Stereobildern},
    Author = {Roscher, Ribana},
    Institution = {Department of Photogrammetry, University of Bonn},
    Year = {2008},
    Number = {TR-IGG-P-2008-05},
    Abstract = {Dieser Report erl\"autert die Bestimmung von 3D-Merkmalen von Bildregionen durch Zuordnung von Bildpunkten in diesen Regionen zu den Objektpunkten in einer Punktwolke.Die Umsetzung erfolgt in einer grafischen Benutzeroberfl\"ache in Matlab, deren Bedienung in diesem Report veranschaulicht werden soll.},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2008Bestimmung.pdf}
    }

  • R. Roscher, “Lernen linearer probabilistischer diskriminativer Modelle für die semantische Bildsegmentierung,” Diploma Thesis Master Thesis, 2008.
    [BibTeX] [PDF]
    [none]
    @MastersThesis{Roscher2008Lernen,
    Title = {Lernen linearer probabilistischer diskriminativer Modelle f\"ur die semantische Bildsegmentierung},
    Author = {Roscher, Ribana},
    School = {Institute of Photogrammetry, University of Bonn},
    Year = {2008},
    Note = {Betreuung: Prof. Dr.-Ing. Wolfgang F\"orstner, Ing. Filip Kor{\vc}},
    Type = {Diploma Thesis},
    Abstract = {[none]},
    Url = {http://www.ipb.uni-bonn.de/pdfs/Roscher2008Lernen.pdf}
    }

igg
Institute of Geodesy
and Geoinformation
lwf
Faculty of Agriculture
ubn
University of Bonn