In this letter, we address the characterization of objects in 3-D sonar images of the water column obtained by a multibeam echo sounder. Compared with classic 2-D images from a monobeam echo sounder, these 3-D images provide finer scale observation of the pelagic biomasses and new tools to characterize 3-D distributions. By viewing object patterns as realizations of spatial point processes, we investigate descriptive spatial statistics. This method is then applied to 3-D fisheries acoustics data set for characterization of the distribution of pelagic fish schools. Reported experiments illustrate the relevance of the proposed descriptors. The comparison of our method with 2-D sonar data analysis further demonstrates the information gain from using 3-D sonar imagery.

In this paper, we address seabed characterization and recognition in sonar images using keypoint-based approaches. Keypoint-based texture recognition has recently emerged as a powerful framework to address invariances to contrast change and geometric distortions. We investigate here to which extent keypoint-based techniques are relevant for sonar texture analysis which also involves such invariance issues. We deal with both the characterization of the visual signatures of the keypoints and the spatial patterns they form. In this respect, spatial statistics are considered. We report a quantitative evaluation for sonar seabed texture data sets comprising six texture classes such as mud, rock, and gravely sand. We clearly demonstrate the improvement brought by keypoint-based techniques compared to classical features used for sonar texture analysis such as cooccurrence and Gabor features. In this respect, we demonstrate that the joint characterization of the visual signatures of the visual keypoints and their spatial organization reaches the best recognition performances (about 97% of correct classification w.r.t. 70% and 81% using cooccurrence and Gabor features). Furthermore, the combination of difference of Gaussian keypoints and scale-invariant feature transform descriptors is recommended as the most discriminating keypoint-based framework for the analysis of sonar seabed textures.

In this paper, we address invariant keypoint-based texture characterization and recognition. Viewing keypoint sets associated with visual textures as realizations of point processes, we investigate probabilistic texture models from multivariate log-Gaussian Cox processes. These models are parameterized by the covariance structure of the spatial patterns. Their implementation initially rely on the construction of a codebook of the visual signatures of keypoints. We discuss invariance properties of the proposed models for texture recognition applications and report a quantitative evaluation for three texture datasets, namely: UIUC, KTH-TIPs and Brodatz. These experiments include a comparison of the performance reached using different methods for keypoint detection and characterization and demonstrate the relevance of the proposed models w.r.t. state-of-the-art methods. We further discuss the main contribution of proposed approach, including the key features of a statistical model and complexity aspects.

In this paper, invariant sonar texture characterization for seabed classification is addressed from the spatial distribution of image keypoints using log-Gaussian Cox processes. Considering the categorized visual keypoints, the spatial statistical properties of keypoint sets are expressed by the intensity and the pair correlation function of the log-Gaussian Cox model to define a novel invariant texture descriptor. Reported results of an application to sonar texture classification validate the proposed descriptor compared to previous work. We further discuss the main contribution of proposed approach, including the key features of a statistical model and complexity aspects.

In this paper, we address invariant scene classification from images. We propose a novel descriptor based on the statistical characterization of the spatial patterns formed by elementary objects in images. Elementary objects are defined from a tree of shapes of the topology map of the image and each object is characterized by shape context feature vector. Viewing the set of elementary objects as a realization of a random spatial process, we investigate a statistical analysis using log- Gaussian Cox model to define an invariant image descriptor. An application to natural scene recognition is described. Re- ported results validate the proposed descriptor with respect to previous work.

This paper addresses the training of classification trees for weakly labelled data. We call “weakly labelled data”, a training set such as the prior labelling information provided refers to vector that indicates the probabilities for instances to belong to each class. Classification tree typically deals with hard labelled data, in this paper a new procedure is suggested in order to train a tree from weakly labelled data. Resulting tree is different than usual in the sense that weak labels are taking into account and affected to test instances. Considering a forest, we show how trees can be associated in the test step. The proposed method is compared with typical models such as generative and discriminative methods for object recognition and we show that our model can outperform the two previous. The considered models are evaluated on standard datasets from UCI and an application to fisheries acoustics is considered.

This paper addresses the development of invariant descriptors for sonar texture based on spatial statistics of local features. We suggest using a hierarchical clustering algorithm to construct a set of codebook from the vector descriptor of keypoints. Spatial point process model allows us to estimate a co-occurrence statistics of the marks of neighboring keypoints in various study region. The resulting descriptor is applied to texture classification using a discriminative method (K-NN or SVM). Experiments were carried out on a set of real sidescan sonar images aiming to compare our proposed descriptor with other texture descriptors.

This paper investigates variational region-level criterion for supervised and unsupervised texture-based image segmentation. The focus is given to the demonstration of the effectiveness and robustness of this region-based formulation compared to most common variational approaches. The main contributions of this global criterion are twofold. First, the proposed methods circumvent a major problem related to classical texture based segmentation approaches. Existing methods, even if they use different and various texture features, are mainly stated as the optimization of a criterion evaluating punctual pixel likelihoods or similarity measure computed within a local neighborhood. These approaches require sufficient dissimilarity between the considered texture features. An additional limitation is the choice of the neighborhood size and shape. These two parameters and especially the neighborhood size significantly influence the classification performances: the neighborhood must be large enough to capture texture structures and small enough to guarantee segmentation accuracy. These parameters are often set experimentally. These limitations are mitigated with the proposed variational methods stated at the region-level. It resorts to an energy criterion defined on image where regions are characterized by nonparametric distributions of their responses to a set of filters. In the supervised case, the segmentation algorithm consists in the minimization of a similarity measure between region-level statistics and texture prototypes and a boundary based functional that imposes smoothness and regularity on region boundaries. In the unsupervised case, the data-driven term involves the maximization of the dissimilarity between regions. The proposed similarity measure is generic and permits optimally fusing various types of texture features. It is defined as a weighted sum of Kullback-Leibler divergences between feature distributions. The optimization of the proposed variational criteria is carried out using a level-set formulation. The effectiveness and the robustness of this formulation at region-level, compared to classical active contour methods, are evaluated for various Brodatz and natural images.

This paper addresses weakly supervised object recognition. We show how the combination of an image-level inference, in terms of image-level object class priors, can lead to better training of object recognition models. Stated within a probabilistic setting, the proposed approach is applied to fisheries acoustics and fish school recognition.

In this paper, we propose and compare two supervised algorithms for the segmentation of textured sonar images, with respect to seafloor types. We characterize seafloors by a set of empirical distributions estimated on texture responses to a set of different filters. Moreover, we introduce a novel similarity measure between sonar textures in this feature space. Our similarity measure is defined as a weighted sum of Kullback-Leibler divergences between texture features. The weight setting is twofold. First, each filter is weighted according to its discrimination power: The computation of these weights are issued from a margin maximization criterion. Second, an additional weight, evaluated as an angular distance between the incidence angles of the compared texture samples, is considered to take into account sonar-image acquisition process that leads to a variability of the backscattered value and of the texture aspect with the incidence-angle range. A Bayesian framework is used in the first algorithm where the conditional likelihoods are expressed using the proposed similarity measure between local pixel statistics and the seafloor prototype statistics. The second method is based on a variational framework as the minimization of a region-based functional that involves the similarity between global-region texture-based statistics and the predefined prototypes.

This paper addresses the inference of probabilistic classification models using weakly supervised learning. In contrast to previous work, the use of proportion-based training data is investigated in combination to non-linear classification models. An application to fisheries acoustics and fish school classification is considered and experiments are reported for synthetic and real datasets.

Backscattered signal statistics are widely used for target detection and seafloor characterization. The${\cal K}$-distribution shows interesting properties for describing experimental backscattered intensity statistics. In addition to the fact that its probability distribution function accurately fits actual sonar data, it advantageously provides a physical interpretation linked to the backscattering phenomenon. Sonar systems usually record backscattered signals from a wide angular range across the ship's track. In this context, previous studies have shown that backscatter statistics strongly depend on the incidence angle. In this paper, we propose an extension of previous works to model the angular evolution of the${\cal K}$-distribution shape parameter. This evolution is made clear and analyzed from experimental data recorded with two sonar systems: a 95-kHz multibeam echosounder and a 110-kHz sidescan sonar. Model fitting with data backscattered from six seafloor configurations shows the improvement provided by our extension as compared to two previous models.

We propose a novel multi-class method for texture segmentation. The segmentation issue is stated as the minimization of a region-based functional that involves a weighted Kullback-Leibler measure between distributions of local texture features and a regularization term that imposes smoothness and regularity of region boundaries. The proposed approach is implemented using level-set methods, and partial differential equations (PDE) are expressed using shape derivative tools introduced in S. Jehan-Besson et al. (2003). As an application, we have tested the method using cooccurrence distributions to segment synthetic mosaics of textures from the Brodatz album, as well as real textured sonar images. These results prove the relevance of the proposed approach for supervised and unsupervised texture segmentation

An effective denoising method for ECG signals affected by real sources of noise is proposed in this paper. The method is based on a maximum a-posteriori (MAP) filtering in the diversity-enhanced wavelet domain, under realistic a-priori assumptions regarding the statistical properties of the wavelet coefficients of the ECG signal. In order to evaluate the performance of the method, we studied the signal-to-noise ratio (SNR) improvement factor and the degree of the denoising influence on the automatic signal segmentation procedures. The method was tested in both synthetic and real noise conditions and it showed very promising results

We propose a technique, based on a fuzzy hidden Markov chain (HMC) model, for the unsupervised segmentation of images. Our main contribution is to use Dirac and Lebesgue measures simultaneously at the class chain level. This model allows the coexistence of hard and fuzzy pixels in the same picture. The fuzzy approach enriches the classical model by adding a fuzzy class, which has several interpretations in signal processing. One such interpretation in image segmentation is the simultaneous appearance of several thematic classes on the same pixel (mixture). Model parameter estimation is performed through an extension of the iterative conditional estimation (ICE) algorithm to take into account the fuzzy part. The fuzzy segmentation of a real image of clouds is studied and compared to the classification obtained with a "classical" hard HMC model.

In seismic deconvolution, blind approaches must be considered in situations where the reflectivity sequence, the source wavelet signal and the noise power level are unknown. In the presence of long, non-minimum-phase, source wavelets, strong interference of the contributions of reflectors make the wavelet estimation and deconvolution procedure from recorded data complicated. We address this problem in a two step approach. First, a robust, but truncated, estimate of the wavelet is performed using a standard maximum likelihood approach. Then, improved wavelet estimation is achieved by fitting an ARMA model to the initial MA wavelet using the Prony algorithm. The algorithmic problem of wavelet initialization is also addressed. Simulation results and real data experiments show that a significant improvement is brought by this approach.

Classification of high resolution SAR images is difficult due to the presence of speckle noise. We propose to use a multiscale decomposition that allows different trade-off between spatial precision (resolution) and radiometric uncertainty (noise reduction). Classification decisions at large scale are certain but spatially imprecise whereas decisions at high resolution are uncertain but spatially precise. We first decompose the SAR image in low and high frequency images at different scales using a stationary wavelet transformation. Then low pass images are classified by maximum likelihood based on a Gaussian mixture estimation. Wavelet coefficients in high frequency images enable us to identify stationary homogeneous regions within the image where classification decisions are expected to be stable across scales. Decisions at different scales are merged using Dempster-Shafer theory which gives us an adequate framework to manipulate both uncertainty and imprecision. Finally, resulting multiscale decisions are injected in a stochastic classification algorithm (MPM) as a hidden "evidential" Markov random field. The proposed algorithm is evaluated on artificial SAR images. We also propose to filter wavelet coefficients based on the resulting multiscale confidence map.

Radar images can show great variability from pixel to pixel, which is an obstacle to effective processing. This variability, due to speckle created by the radar wave coherence, necessitates the use of more adapted filters. Previous studies have shown that multiresolution wavelet analysis yields better results but produces artefacts due to multiscale decomposition. This paper proposes a method that reduces these effects by introducing the fractal dimension. The resultant filter combines wavelet decomposition and variance change model based on the level of variance estimated by studying the fractal dimension of the image.

A new method of multiband image fusion is proposed. The algorithm, based on an unsupervised method, takes the spatial contextual information into account in the application of Dempster-Shafer theory and sets up initialization and regularization processes adapted to remote sensing images. The importance of introducing spatial contextual information has been illustrated through forest area classification using the Landsat TM sensor.

This paper presents a joint speech and audio coding algorithm combining sinusoidal modeling and a perceptually adapted Wavelet Packet Transform (WPT). The input signal is limited to the band of 50-7000 Hz, and sampled at 16 kHz. The sinusoidal modeling uses a Sinusoidal Similarity Measure(SSM) to find stable sinusoidal components. A novel pitch harmonics based encoding is applied to encode the sinusoidal frequencies. The residual is obtained by extracting the re-synthesized sinusoids from the input, and is processed by a WPT simulating the critical bands of the Human Auditory System . Perceptual Noise Substitution (PNS) is applied in noisy WPT sub-bands to reduce the bit rate. The method provides nearly transparent quality for both speech and audio inputs. The mean bit rate of the compressed signal varies between 32-62 kbps depending on the input. Demonstration sound files are available at www-sc.enstbretagne.fr/˜fek/eurospeech01.

Synthetic aperture radar (SAR) images are disturbed by a multiplicative noise depending on the signal (the ground reflectivity) due to the radar wave coherence. Images have a strong variability from one pixel to another reducing essentially the efficiency of the algorithms of detection and classification. We propose to filter this noise with a multiresolution analysis of the image. The wavelet coefficient of the reflectivity is estimated with a Bayesian model, maximizing the a posteriori probability density function. The different probability density function are modeled with the Pearson system of distributions. The resulting filter combines the classical adaptive approach with wavelet decomposition where the local variance of high-frequency images is used in order to segment and filter wavelet coefficients.

We propose a Bayesian approach for multipath channel time delay estimation when the signal is received by a square-law detector. Two different methods are presented. The first one is applied when the signal is supplied at the output of the detector. On certain approximations, the derivation of the MAP (maximum a posteriori) estimator leads to optimize a l/sub 1/-norm criterion. In the second method, the signal is observed after demodulation and matched-filtering. We introduce a Bernoulli-Gaussian model to account for the sparse properties of the channel impulse response and solve the deconvolution problem using Monte-Carlo simulations. Comparisons between both algorithms are presented on simulations. They show that the second method leads to more accurate estimators, while justifying the validity of the approximations used in the first one.

The problem of wavelet estimation for marine seismology is investigated with a Bayesian approach applied to a Bernoulli-Gaussian model. We specify proper prior distributions for all unknown quantities including the seismic wavelet, the parameters of the reflectivity sequence and noise. To solve this estimation problem, an algorithm close to a stochastic version of the EM algorithm is used. The random variables are generated iteratively by a simple Monte-Carlo method namely the Gibbs sampler. But the direct application of this procedure often leads to a local minimum of the likelihood function resulting in a shifted and distorted wavelet. We propose a general method to obtain the true solution which systematically uses different shifted wavelets to reinitialize the algorithm. Then we rerun the procedure on each initialization and retain the wavelet which minimizes the noise variance.

We propose the use of evidential reasoning in order to relax the Bayesian decisions given by a multiscale Markovian classification algorithm (ICM). The Dempster-Shafer combination rule enables us to fuse decisions in a local spatial neighbourhood which we further extend to be multiscale and multisource. This approach enables us to more directly fuse multiscale information. Application to the classification of very noisy radar images has produced interesting results.

Radar images are affected by multiplicative noise depending on the underlying signal (the ground reflectivity) due to the coherence of the radar wavelength. Images present a strong pixel to pixel variability considerably reducing the efficiency of target detection and classification algorithms. We propose in this study filtering this noise using image multiresolution analysis. The value of the wavelet coefficients of the radar reflectance is estimated by a Bayesian model by maximizing the a posteriori density and by modeling the different densities using the Pearson distributions system. The resulting filter combines a classical adaptive approach and wavelet decomposition using the local variance of the wavelet coefficients for segmenting and weighting the latter taking into account the multiplicative nature of the noise.

International audience

In this paper a new speckle filter for SAR images is proposed, based on a wavelet decomposition. The N-levels decomposition of the initial image with a filter bank provides high frequency images containing wavelet coefficients and low frequency images. This analysis tool allows us to separate details, edges, ... from low frequency information (intensity averages). High frequency images are treated specifically in order to keep only significant wavelet coefficients. An adaptive filter is applied on the low frequency images during the reconstruction stage so as to eliminate the residual speckle.

Presents the development of the RGSKA/spl epsi/, a new algorithm for designing vector quantizers. The main features of this algorithm are the following: due to its stochastic nature it avoids being trapped in poor local minima; initial codebook is not needed; the codevectors move away from the gravity centre of the training vectors towards their final position; source coding and channel coding are jointly optimized to obtain a codebook robust against different levels of the transmission noise; The resulted codebook always performs as well or even better than existing codebooks designed for noisy or noiseless channels; the computational complexity is only slightly higher than that of the most widely used K-means algorithm; the bootstrap sampling technique can be successfully applied in case of a large training set; the method is suitable for parallel implementation.

International audience

Intensity radar images are difficult to classify because of the speckle phenomenon, which acts like a multiplicative noise and which is characterized by a Gamma distribution law. Unsupervised Bayesian segmentation applied to the whole radar image gives only good results in terms of classification rate when the look number is sufficiently high to approximate the Gamma law by a Gaussian law [1]. Multiresolution image analysis by wavelets has proved to be efficient for increasing classification performances in this Gaussian case [2,3], and also for optical images [4]. In this paper, it is proposed to extend the method to nongaussian images,by using cumulants to approximate the conditional distribution used in the classification algorithm at each level of the pyramid and to apply the procedure to simulated radar images with a low look number.>

Deals with unsupervised Bayesian segmentation of images. The authors introduce a new algorithm based on a recent general method of estimation in the case of incomplete data (iterative conditional estimation). The efficiency of the method is compared with a recent algorithm based on the stochastic gradient by L. Younes (1989). Results of numerous simulations are given and an application to a real radar image is also derived.>

In a surveillance system in an aerospace environment which has the task of detecting a source a long distance away and thus has a very low resolution, cloudy sky velocity estimation should allow a lower rate of false alerts in discriminating the motion between the different patterns. This kinematic field estimation with little apparent motion could be effective by using the generalized Hough transform method or with a differential method based on optical flow by which one should locally obtain the kinematic field estimation. The paper aims at showing and adapting this differential method in areas where there is motion. These areas are detected by means of an entropy criterion. The method can discern the global movement of objects whatever their form. A local kinematic vector estimation method, based on the least mean squares is then tested.>

International audience

International audience

International audience

Automatic identification and counting of zooplankton is becoming very useful in marine ecology to avoid long and tedious tasks for biologists. We propose a method of image analysis to detect edges, filter the binary image to reduce spots and appendices such as legs or antennas, and measure parameters, such as the length of the abdomen which are finally used in the classification. Some results of classification on 320 specimens are then presented.