Showing papers by "Jocelyn Chanussot published in 2009"

Spectral–Spatial Classification of Hyperspectral Imagery Based on Partitional Clustering Techniques

Abstract: A new spectral-spatial classification scheme for hyperspectral images is proposed. The method combines the results of a pixel wise support vector machine classification and the segmentation map obtained by partitional clustering using majority voting. The ISODATA algorithm and Gaussian mixture resolving techniques are used for image clustering. Experimental results are presented for two hyperspectral airborne images. The developed classification scheme improves the classification accuracies and provides classification maps with more homogeneous regions, when compared to pixel wise classification. The proposed method performs particularly well for classification of images with large spatial structures and when different classes have dissimilar spectral responses and a comparable number of pixels.

Kernel principal component analysis for the classification of hyperspectral remote sensing data over urban areas

Abstract: Kernel principal component analysis (KPCA) is investigated for feature extraction from hyperspectral remote sensing data. Features extracted using KPCA are classified using linear support vector machines. In one experiment, it is shown that kernel principal component features are more linearly separable than features extracted with conventional principal component analysis. In a second experiment, kernel principal components are used to construct the extended morphological profile (EMP). Classification results, in terms of accuracy, are improved in comparison to original approach which used conventional principal component analysis for constructing the EMP. Experimental results presented in this paper confirm the usefulness of the KPCA for the analysis of hyperspectral data. For the one data set, the overall classification accuracy increases from 79% to 96% with the proposed approach.

Decision Fusion for the Classification of Hyperspectral Data: Outcome of the 2008 GRS-S Data Fusion Contest

Abstract: The 2008 Data Fusion Contest organized by the IEEE Geoscience and Remote Sensing Data Fusion Technical Committee deals with the classification of high-resolution hyperspectral data from an urban area. Unlike in the previous issues of the contest, the goal was not only to identify the best algorithm but also to provide a collaborative effort: The decision fusion of the best individual algorithms was aiming at further improving the classification performances, and the best algorithms were ranked according to their relative contribution to the decision fusion. This paper presents the five awarded algorithms and the conclusions of the contest, stressing the importance of decision fusion, dimension reduction, and supervised classification methods, such as neural networks and support vector machines.

Pansharpening Quality Assessment Using the Modulation Transfer Functions of Instruments

Abstract: Quality assessment of pansharpening methods is not an easy task. Quality-assessment indexes, like Q4, spectral angle mapper, and relative global synthesis error, require a reference image at the same resolution as the fused image. In the absence of such a reference image, the quality of pansharpening is assessed at a degraded resolution only. The recently proposed index of Quality Not requiring a Reference (QNR) is one among very few tools available for assessing the quality of pansharpened images at the desired high resolution. However, it would be desirable to cross the outcomes of several independent quality-assessment indexes, in order to better determine the quality of pansharpened images. In this paper, we propose a method to assess fusion quality at the highest resolution, without requiring a high-resolution reference image. The novel method makes use of digital filters matching the modulation transfer functions (MTFs) of the imaging-instrument channels. Spectral quality is evaluated according to Wald's spectral consistency property. Spatial quality measures interscale changes by matching spatial details, extracted from the multispectral bands and from the panchromatic image by means of the high-pass complement of MTF filters. Eventually, we highlight the necessary and sufficient condition criteria for quality-assessment indexes by developing a pansharpening method optimizing the QNR spatial index and assessing the quality of fused images by using the proposed protocol.

Support Vector Reduction in SVM Algorithm for Abrupt Change Detection in Remote Sensing

Abstract: Satellite imagery classification using the support vector machine (SVM) algorithm may be a time-consuming task. This may lead to unacceptable performances for risk management applications that are very time constrained. Hence, methods for accelerating the SVM classification are mandatory. From the SVM decision function, it can be noted that the classification time is proportional to the number of support vectors (SVs) in the nonlinear case. In this letter, four different algorithms for reducing the number of SVs are proposed. The algorithms have been tested in the frame of a change detection application, which corresponds to a change-versus-no-change classification problem, based on a set of generic change criteria extracted from different combinations of remote sensing imagery.

On the use of ICA for hyperspectral image analysis

Abstract: Independent component analysis (ICA) is a very popular method that has shown success in blind source separation, feature extraction and unsupervised recognition. In recent years ICA has been largely studied by researchers from the signal processing community. This paper addresses a more in-depth study on the use of this method, applied to hyper-spectral images used for remote sensing purposes. In a first part, source separation is addressed. Since the independence of sources is usually not verified in hyperspectral real data images, ICA, if used alone, is not a suitable tool to unmix sources. We propose a hierarchical approximation for the use of ICA as a pre-processing step for a Bayesian Positive Source Separation method. In a second part, the use of ICA for dimensionality reduction is studied in the frame of hyperspectral data classification. Experimental results show the effectiveness of ICA when used for hyperspectral image pre-processing for the two considered applications.

Machine Learning Techniques in Remote Sensing Data Analysis

Abstract: Several applications have been developed in the field of remote sensing image analysis during the last decades. Besides well-known statistical approaches, many recent methods are based on techniques taken from the field of machine learning. A major aim of machine learning algorithms in remote sensing is supervised classification, which is perhaps the most widely usedimageclassificationapproach.Inthischapterabriefintroductiontomachinelearningand the different paradigms in remote sensing is given. Moreover this chapter briefly discusses the use of recent developments in supervised classification techniques such as neural networks, support vector machines and multiple classifier systems.

Multiscale stochastic watershed for unsupervised hyperspectral image segmentation

Abstract: This paper deals with unsupervised segmentation of hyper-spectral images. It is based on the stochastic watershed, an approach to estimate a probability density function (pdf) of contours of an image using MonteCarlo simulations of watershed segmentations. In particular, it is introduced for the first time a multiscale framework for the computation of the pdf of contours using the stochastic watershed. Two multiscale approaches are considered: i) a linear scale-space using Gaussian filters, ii) a nonlinear morphological scale-space pyramid using levelings. In addition, a multiscale pyramid obtained by modifying the size of the random markers is also studied. Then, it is shown how the pdf of contours can finally be segmented using the non-parametric waterfalls algorithm. The performances of the proposed methods are compared using two examples of standard remote sensing hyperspectral images.

Unsupervised classification of hyperspectral images by using linear unmixing algorithm

Abstract: In this paper, we present an unsupervised classification algorithm for hyperspectral images. For reducing the dimension of hyperspectral data, we use a linear unmixing algorithm to extract the endmembers and their abundance maps. Compared to the components obtained by traditional PCA-basedmethod, the abundancemaps have physical meanings (such as the abundance of vegetation). For determining the number of endmembers contained in an image, we propose an eigenvalue based approach. The validation of this approach on synthetic data shows that this approach provides a robust estimation of the actual number of endmembers. Using the estimated abundance maps of the endmemebers, we perform a preliminary segmentation and use the mean values of the segmented regions as feature for the classification. We then perform Kmeans classifications on the segmented abundance maps with the number of clusters determined by the Krzanowski and Lai's method.

Multivariate Image Processing

Abstract: 1. Introduction to Multivariate Image Processing from the Basics to New Challenges. 2. Registration. 3. Fusion of SAR and Optical Data. 4. Fusion of Satellite Images at Different Resolutions. 5. Multitemporal Processing and Change Detection. 6. Bayesian Approach to Linear Spectral Mixture Analysis. 7. Detection and Tracking of Emission Rays in Radioastronomy. 8. Wavelet Transform for the Denoising of Multivariate Images. 9. Bayesian Approach for Polarization-encoded Image Analysis. 10. Unsupervised Classification for Multivariate Images. 11. Noise Estimation.

Classification based marker selection for watershed transform of hyperspectral images

Abstract: A new method for segmentation and classification of hyper-spectral images is proposed. The method is based on a pixel-wise classification followed by selection of the most reliable classified pixels as markers for watershed segmentation. Furthermore, each marker defined from classification results is associated with a class label. By assigning the class label of each marker to all the pixels within the region grown from this marker, a spectral-spatial classification map is obtained. Experimental results are presented on a 200-band AVIRIS image of the Northwestern Indiana's Indian Pine site. The developed segmentation and classification scheme significantly decreases oversegmentation, improves classification accuracies and provides classification maps with more homogeneous regions, when compared to pixel-wise classification or previously proposed spectral-spatial classification techniques.

Hyperspectral image classification based on spectral and geometrical features

Abstract: In this paper, we propose to integrate geometrical features, such as the characteristic scales of structures, with spectral features for the classification of hyperspectral images The spectral features which only describe the material of structures can not distinguish objects made by the same material but with different semantic meanings (such as the roofs of some buildings and the roads) The use of geometrical features is therefore necessary Moreover, since the dimension of a hyperspectral image is usually very high, we use linear unmixing algorithm to extract the endmemebers and their abundancemaps in order to represent compactly the spectral information Afterwards, with the help of these abundance maps, we propose a method based on topographic map of images to estimate local scales of structures in hyperspectral images The experiment shows that the geometrical features can improve the classification results, especially for the classes made by the same material but with different semantic meanings When compared to the traditional contextual features (such as morphological profiles), the local scale feature provides satisfactory results without considerably increasing the feature dimension

Pansharpening of Hyperspectral images using spatial distortion optimization

Abstract: This paper presents a novel method for the spatial quality improvement of low resolution Hyperspectral (HS) images by making use of a high resolution panchromatic (Pan) image. Since the introduction of all the details extracted from the Pan image into the upscaled HS images may result in spectral and spatial distortions, a detail injection model based on the optimization of QNR spatial quality index is proposed. This proposed model produces pansharpened images while preserving spectral fidelity. Also, to speed up the fusion process we propose to use the Universal Image Quality Index (UIQI) for dimensionality reduction before performing pansharpening. Finally, a comparison of the proposed method is presented with some existing pansharpening methods.

Morphological image distances for hyperspectral dimensionality exploration using Kernel-PCA and ISOMAP

Abstract: MOTIVATION AND AIM The application of nonlinear manifold learning for hyperspec-tral image analysis has been widely studied in last years [1, 4]. One of the main ingredients of these data reduction techniques is the distance used to compare the spectral band images. By means of this distance the pairwise similarity matrix is built and then, the matrix is used to explore the intrinsic dimensionality of the hyperspectral image.

Directional mathematical morphology for the detection of the road network in Very High Resolution remote sensing images

Abstract: This paper presents a new method for extracting roads in Very High Resolution remotely sensed images based on advanced directional morphological operators. The proposed approach introduces the use of Path Openings and Closings in order to extract structural pixel information. These morphological operators remain flexible enough to fit rectilinear and slightly curved structures since they do not depend on the choice of a structural element shape and hence outperform standard approaches using rotating rectangular structuring elements. The method consists in building a granulometry chain using Path Openings and Closing to perform Morphological Profiles. For each pixel, the Morphological Profile constitutes the feature vector on which our road extraction is based.

Hyperspectral pansharpening using QNR optimization constraint

Abstract: This paper presents a method for pansharpening of low resolution Hyperspectral (HS) images. The proposed method is based upon the optimization of both the spectral and spatial quality criteria of the QNR quality assessment index. The simultaneous optimization of the spectral and spatial quality constraints is obtained by means of the Pareto solutions, obtained by making use of an evolutionary algorithm. A selection criteria is defined to select a single solution from among the Pareto solutions and the results obtained show both quantitative and qualitative improvement over the results obtained by some existing pansharpening methods.

Detection and localization of transient sources: comparative study of complex-lag distribution concept versus wavelets and spectrogram-based methods

Abstract: The detection and localization of transient signals is nowadays a typical point of interest when we consider the multitude of existing transient sources, such as electrical and mechanical systems, underwater environments, audio domain, seismic data, and so forth. In such fields, transients carry out a lot of information. They can correspond to a large amount of phenomena issued from the studied problem and important to analyze (anomalies and perturbations, natural sources, environmental singularities, ...). They usually occur randomly as brief and sudden signals, such as partial discharges in electrical cables and transformers tanks. Therefore, motivated by advanced and accurate analysis, efficient tools of transients detection and localization are of great utility. Higher order statistics, wavelets and spectrogram distributions are well known methods which proved their efficiency to detect and localize transients independently to one another. However, in the case of a signal composed by several transients physically related and with important energy gap between them, the tools previously mentioned could not detect efficiently all the transients of the whole signal. Recently, the generalized complex time distribution concept has been introduced. This distribution offers access to highly concentrated representation of any phase derivative order of a signal. In this paper, we use this improved phase analysis tool to define a new concept to detect and localize dependant transients taking regard to the phase break they cause and not their amplitude. ROC curves are calculated to analyze and compare the performances of the proposed methods.

Ensemble Strategies for Classifying Hyperspectral Remote Sensing Data

Abstract: The classification of hyperspectral imagery, using multiple classifier systems is discussed and an SVM-based ensemble is introduced. The data set is separated into separate feature subsets using the correlation between the different spectral bands as a criterion. Afterwards, each source is classified separately by an SVM classifier. Finally, the different outputs are used as inputs for final decision fusion that is based on an additional SVM classifier. The results using the proposed strategy are compared to classification results achieved by a single SVM and other well known classifier ensembles, such as random forests, boosting and bagging.

Classification of hyperspectral images using automatic marker selection and Minimum Spanning Forest

Abstract: A new method for segmentation and classification of hyperspectral images is proposed. The method is based on the construction of a Minimum Spanning Forest (MSF) from region markers. Markers are defined automatically from classification results. For this purpose, pixel-wise classification is performed and the most reliable classified pixels are chosen as markers. Furthermore, each marker defined from classification results is associated with a class label. Each tree in the MSF grown from a marker forms a region in the segmentation map. By assigning a class of each marker to all the pixels within the region grown from this marker, classification map is obtained. Furthermore, the classification map is refined, using results of a pixel-wise classification and a majority voting within the spatially connected regions. Experimental results are presented on a 200-band AVIRIS image of the Northwestern Indiana's Indian Pine site. The use of different dissimilarity measures for construction of the MSF is investigated. The proposed scheme improves classification accuracies, when compared to previously proposed classification techniques, and provides accurate segmentation and classification maps.

Unsupervised endmember extraction: Application to hyperspectral images from Mars

Abstract: In this paper, we try to identify and quantify the chemical species present on the surface of planet Mars with the help of hyperspectral images provided by the instrument OMEGA. For this purpose, we suppose that the spectrum of each pixel is a linear mixture of the spectra of different endmembers. From this linear mixture hypothesis, our work is divided into two steps. Firstly, we propose a new unsupervised method for estimating the number of endmembers based on the eigenvalues of covariance and correlation matrix of the hyperspectral data. This method is then validated on synthetic data. With the help of the number estimated by the precedent step, we use the Vertex Component Analysis (VCA) to extract the spectra and the abundances of the endmembers. The results on hyperspectral image taken by the instrument OMEGA are shown.

Spectral-spatial analysis in hyperspectral remote sensing: from morphological profiles to classified segmentation

Abstract: In this paper, we cover a decade of research in the field of spectral-spatial classification in hyperspectral remote sensing. While the very rich spectral information is usually used through pixel-wise classification in order to recognize the physical properties of the sensed material, the spatial information, with a constantly increasing resolution, provides insightful features to analyze the geometrical structures present in the picture. This is especially important for the analysis of urban areas, while this helps reducing the classification noise in other cases. The very high dimension of hyperspectral data is a very challenging issue when it comes to classification. Support Vector Machines are nowadays widely aknowledged as a first choice solution. In parallel, catching the spatial information is also very challenging. Mathematical morphology provides adequate tools: granulometries (the morphological profile) for feature extraction, advanced filters for the definition of adaptive neighborhoods, the following natural step being an actual segmentation of the data. In order to merge spectral and spatial information, different strategies can be designed: data fusion at the feature level or decision fusion combining the results of a segmentation on the one hand and the result of a pixel wise classification on the other hand.

Unsupervised endmember extraction of martian hyperspectral images

Abstract: In this paper, we try to identify and quantify the chemical species present on the surface of planet Mars with the help of hyperspectral images provided by the instrument OMEGA [1]. For this purpose, we suppose that the spectrum of each pixel is a linear mixture of the spectra of different endmembers. From this linear mixture hypothesis, our work is divided into two steps. Firstly, we propose a new unsupervised method for estimating the number of endmembers based on the eigenvalues of covariance and correlation matrix of the hyperspectral data. This method is then validated on synthetic data. With the help of the number estimated by the precedent step, we use the Vertex Component Analysis (VCA) to extract the spectra and the abundances of the endmembers. The results on hyperspectral image acquired by the OMEGA instrument are shown.

Détection automatique du réseau vasculaire rétinien basée sur la morphologie directionnelle et la fusion de décision

Abstract: In this paper, we address the problem of detecting blood vessels in retinal fundus images. We propose a method based on the construction of vessel features vectors to perfom a pixel classification between pixels belonging or not belonging to the vessel tree. A vessel feature extraction is described using advanced morphological directional filters, Path Openings. The proposed vessels features are linked to the contrast of vessels and their linear connectivity. A data fusion of these features, based on fuzzy set theory, provides a pixel classification and hence the vessels maps. Experimental results have demonstrated the ability of the proposed method to successfully extract a good quality vessel tree.

Signal characterization using generalized “Time-phase derivatives” representation

Abstract: Natural signals are often characterized by a complex time-frequency behaviour. These signals exist in many different applications and systems from underwater acoustic to audio signals with sound attacks or electrical systems with partial discharges and commutation switches, for example. There is a huge number of Time-Frequency (TF) methods that aim to characterize these signals in terms of first phase derivative analysis (i.e Instantaneous Frequency Law). Recently, we introduced the time frequency distribution based on complex lag arguments. This distribution is able to reduce inner interferences terms which appear when studying non-linear TF components. It also offers access to an instantaneous law representation of any phase derivative order. In this paper, we use these two properties to study highly non-stationary signals as well as transient signals.

Éditorial - Télédétection pour la surveillance et la gestion de l’environnement

Abstract: Résumé Basée sur une excellence historique dans les domaines du traitement du signal et des images et s’appuyant sur des leaders institutionnels et industriels de l’aéronautique et de l’exploration spatiale (lancement de satellites, développement de capteurs...), la communauté française de télédétection est l’une des plus actives au monde. Accompagnant la création du Chapitre Français de la société IEEE Géoscience et Télédétection (IEEE Geoscience and Remote Sensing Society), ce numéro spécial vise à mettre en lumière des développements méthodologiques récents pour le traitement et l’analyse des signaux et images de télédétection pour la surveillance et la gestion de l’environnement. Ce thème est particulièrement d’actualité, d’une part parce que ces aspects sont au coeur des enjeux sociétaux du 21ème siècle, d’autre part parce que les capteurs actuels permettent désormais effectivement d’observer ces phénomènes avec précision, à moidre coût et sur une grande échelle. Néanmoins, la mise en correspondance des potentialités offertes par les nouveaux capteurs avec les applications envisagées nécessite des développements en traitement du signal. Les articles retenus pour ce numéro spécial s’inscrivent dans cette optique.

Kernel principal component analysis for the construction of the extended morphological profile

Abstract: Kernel Principal Component Analysis (KPCA) is investigated for feature extraction from hyperspectral remote-sensing data. Features extracted using KPCA are used to construct the Extended Morphological Profile (EMP). Classification results, in terms of accuracy, are improved in comparison to original approach which used conventional principal component analysis for constructing the EMP. Experimental results presented in this paper confirm the usefulness of the KPCA for the analysis of hyperspectral data. The overall classification accuracy increases from 79% to 96% with the proposed approach.

From Statistical Detection to Decision Fusion: Detection of Underwater Mines in High Resolution SAS Images

Abstract: Among all the applications proposed by sonar systems is underwater demining. Indeed, even if the problem is less exposed than the terrestrial equivalent, the presence of underwater mines in waters near the coast an d particularly the harbours provoke accidents and victims in fishing and trade activiti es, even a long time after conflicts. As for terrestrial demining (Milisavljeviþ et al., 2008), detection and classification of various types of underwater mines is cu rrently a crucial strategic task (U.S. Department of the Navy, 2000). Over the past decade, synthetic aperture sonar (SAS) has been increasingly used in seabed imaging, providing high-resolution images (Hayes & Gough, 1999). However, as with any active coherent imaging system, the speckle constructs images with a strong granular aspect that can seriously handicap the interpre tation of the data (Abbot & Thurstone, 1979). Many approaches have been proposed in underwater mine detection and classification using sonar images. Most of them use the charac teristics of the shadows cast by the objects on the seabed (Mignotte et al., 1997). These methods fail in case of buried objects, since no shadow is cast. That is why this last case has been less studied. In such cases, the echoes (high-intensity reflection of the wave on the objects) are the only hint suggesting the presence of the objects. Their small size, even in SAS imaging, and the similarity of their amplitude with the background make the detection more complex. Starting from a synthetic aperture image, a complete detection and classification process would be composed of three main parts as follows: 1. Pixel level: the decision consists in deciding whether a pixel belongs to an object or to the background. 2. Object level: the decision concerns the segmented object which is “real” or not: are these objects interesting (mines) or simple rocks, wastes? Shape parameters (size, ) and position information can be us ed to answer this question. 3. Classification of object: the decision concerns the type of object and its identification (type of mine). This chapter deals with the first step of this proc ess. The goal is to evaluate a confidence that a pixel belongs to a sought object or to the seabed. In the following, considering the object

Hyperspectralpansharpeningusing qnr optimizationconstraint

Abstract: This paper presents a method for pansharpening of low res­ olution Hyperspectral (HS) images. The proposed method is based upon the optimization of both the spectral and spatial quality criteria of the QNR quality assessment index. The simultaneous optimization of the spectral and spatial quality constraints is obtained by means of the Pareto solutions, ob­ tained by making use of an evolutionary algorithm. A selec­ tion criteria is defined to select a single solution from among the Pareto solutions and the results obtained show both quan­ titative and qualitative improvement over the results obtained by some existing pansharpening methods.