Dense image matching aims to find a correspondent for every pixel of a source image in a partially overlapping target image. State-of-the-art methods typically rely on a coarse-to-fine mechanism where a single correspondent hypothesis is produced per source location at each scale. In challenging cases – such as at depth discontinuities or when the target image is a strong zoom-in of the source image – the correspondents of neighboring source locations are often widely spread and predicting a single correspondent hypothesis per source location at each scale may lead to erroneous matches. In this paper, we investigate the idea of predicting multiple correspondent hypotheses per source location at each scale instead. We consider a beam search strategy to propagate multiple hypotheses at each scale and propose integrating these multiple hypotheses into cross-attention layers, resulting in a novel dense matching architecture called BEAMER. BEAMER learns to preserve and propagate multiple hypotheses across scales, making it significantly more robust than state-of-the-art methods, especially at depth discontinuities or when the target image is a strong zoom-in of the source image. Our code will be made publicly available.

Variational Autoencoders have gained considerable attention due to their capacity of encoding high dimensional data into a lower dimensional latent space. In this context, several methods have been proposed with the objective of producing disentangled representations. In this work, we propose a weakly supervised model that explicitly disentangles the factors of variation of a dataset in separate subspaces using a pairwise architecture. We also create a framework that encourages conditional image generation according to the desired factor of variation, by controlling these subspaces. This is achieved by introducing an additional network trained with a triplet loss. Its output approximates representations of images generated from the same factor and push the ones of images generated from different factors apart. Experiments are carried out on widely used datasets, and show that our model is able to disentangle specified factors of variation, and to generate new data while constraining desired properties, even when these factors have small influence on reconstruction loss.

Compared to standard deep convolutional neural networks (CNN) which include a global average pooling operator, second-order neural networks have a global covariance pooling operator which allows to capture richer statistics of CNN features. They have been shown to improve representation and generalization abilities. However, this covariance pooling is performed only on the deepest CNN feature maps. To benefit from different levels of abstraction, we propose to extend these models by using a multi-layer approach. In addition, to obtain better predictive performance, an end-to-end ensemble learning architecture is proposed. Experiments are conducted on four datasets and have confirmed the potential of the proposed model for various image processing applications such as remote sensing scene classification, indoor scene recognition and texture classification.

The pansharpening problem amounts to fusing a high-resolution panchromatic image with a low-resolution multispectral image so as to obtain a high-resolution multispectral image. Therefore, the preservation of the spatial resolution of the panchromatic image and the spectral resolution of the multispectral image is of key importance for the pansharpening problem. To cope with it, we propose a new method based on a bidiscriminator in a generative adversarial network (GAN) framework. The first discriminator is optimized to preserve textures of images by taking as input the luminance and the near-infrared band of images, and the second discriminator preserves the color by comparing the chroma components Cb and Cr. Thus, this method allows to train two discriminators, each one with a different and complementary task. Moreover, to enhance these aspects, the proposed method based on bidiscriminator, and called MDSSC-GAN SAM, considers a spatial and a spectral constraint in the loss function of the generator. We show the advantages of this new method on experiments carried out on Pléiades and World View 3 satellite images.

The launch of the last generation of Earth observation satellites has yield to a great improvement in the capabilities of acquiring Earth surface images, providing series of multitemporal images. To process these time series images, many machine learning algorithms have been proposed in the literature such as warping based methods and recurrent neural networks (LSTM,...). Recently, based on an ensemble learning approach, the time series cluster kernel (TCK) has been proposed and has shown competitive results compared to the state-of-the-art. Unfortunately, it does not model the spectral/spatial dependencies. To overcome this problem, this paper aims at extending the TCK approach by modeling the time series of second-order statistical features (SO-TCK). Experimental results are conducted on different benchmark datasets, and land cover classification with remote sensing satellite time series over the Reunion Island.

The pansharpening problem consists in fusing a high resolution panchromatic image with a low resolution multispectral image in order to obtain a high resolution multispectral image. In this paper, we adapt a Residual Dense architecture for the generator in a Generative Adversarial Network framework. Indeed, this type of architecture avoids the vanishing gradient problem faced when training a network by re-injecting previous information thanks to dense and residual connections. Moreover, an important point for the pansharpening problem is to preserve the geometry of the image. Hence, we propose to add a regularization term in the loss function of the generator: it preserves the geometry of the target image so that a better solution is obtained. In addition, we propose geometrical measures that illustrate the advantages of this new method.

Most of existing superpixel methods are designed to segment standard planar images as pre-processing for computer vision pipelines. Nevertheless, the increasing number of applications based on wide angle capture devices, mainly generating 360° spherical images, have enforced the need for dedicated superpixel approaches. In this paper, we introduce a new superpixel method for spherical images called SphSPS (for Spherical Shortest Path-based Superpixels). Our approach respects the spherical geometry and generalizes the notion of shortest path between a pixel and a superpixel center on the 3D spherical acquisition space. We show that the feature information on such path can be efficiently integrated into our clustering framework and jointly improves the respect of object contours and the shape regularity. To relevantly evaluate this last aspect in the spherical space, we also generalize a planar global regularity metric. Finally, the proposed SphSPS method obtains significantly better performances than both planar and spherical recent superpixel approaches on the reference 360° spherical panorama segmentation dataset.

Most superpixel algorithms compute a trade-off between spatial and color features at the pixel level. Hence, they may need fine parameter tuning to balance the two measures, and highly fail to group pixels with similar local texture properties. In this paper, we address these issues with a new Texture-Aware SuperPixel (TASP) method. To accurately segment textured and smooth areas, TASP automatically adjusts its spatial constraint according to the local feature variance. Then, to ensure texture homogeneity within superpixels, a new pixel to super-pixel patch-based distance is proposed. TASP outperforms the segmentation accuracy of the state-of-the-art methods on texture and also natural color image datasets.

This paper presents a new method to perform unsupervised band selection (UBS) with hyperspectral data. The method provides a probabilistic clustering approach. The band images are clustered in the image space by computing their posterior class probability. Then, for each cluster, the band exhibiting the highest probability of belonging to it is selected as cluster exemplar. More particularly, the proposed method falls into information-maximization clustering methods, where the posterior class probability is modeled and the parameters of the models are derived by maximizing the information between the data and the unknown cluster labels. In this context, we propose a new image representation for hyperspectral images, based on the first- and second-order statistics of multiple image features. We refer to this representation as multiple-feature local statistical descriptors (MLSD). The descriptors are computed with respect to regular grids, and a special pixel selection procedure reduces the number of samples within each block of the grid. A kernel-based model that embeds the MLSD is then proposed for the posterior class probability. The model is finally optimized according to an information-maximization criterion. We conduct several experiments to determine the best parameters for the proposed approach and compare the latter with other state-of-the-art UBS methods. Quantitative evaluations show that, by employing our band selection method, higher performance in terms of classification accuracy and endmember extraction can be achieved in comparison with the state of the art.

The most common scheme for estimating the orientation field of space-varying directional textures is based on a local nonlinear spatial averaging of the gradient field. This leads to locally biased orientations, especially in regions of nonlinearly distributed convergence, asymmetrically distributed curvature, or geometry superposition. In this paper, we propose an orientation estimation framework that is invariant toward the local geometry. Instead of applying the spatial averaging in the initial space, it is performed in a flattened space which is obtained from a parametric space transformation model deduced from the reconstruction of local hypersurfaces. The flattening can either be applied to the initial gradient field or to the intensity image, on which the flattened gradient field is computed. The process is iterated in order to define more accurate space transformations leading to refined orientation estimation. The computational efficiency is improved without noticeable loss of orientation accuracy by a patch-based approach. Experiments, both on synthetic as well as real-life fingerprint and fibrous material images, exhibit enhanced performance in comparison with conventional methods such as the structure tensor method.

Many signal and image processing applications, including SAR polarimetry and texture analysis, require the classification of complex covariance matrices. The present paper introduces a geometric learning approach on the space of complex covariance matrices based on a new distribution called Riemannian Gaussian distribution. The proposed distribution has two parameters, the centre of mass Y̅ and the dispersion parameter σ. After having derived its maximum likelihood estimator and its extension to mixture models, we propose an application to texture recognition on the VisTex database.

This paper presents a novel framework for visual content classification using jointly local mean vectors and covariance matrices of pixel level input features. We consider local mean and covariance as realizations of a bivariate Riemannian Gaussian density lying on a product of submanifolds. We first introduce the generalized Mahalanobis distance and then we propose a formal definition of our product-spaces Gaussian distribution on Rm× SPD(m). This definition enables us to provide a mixture model from a mixture of a finite number of Riemannian Gaussian distributions to obtain a tractable descriptor. Mixture parameters are estimated from training data by exploiting an iterative Expectation-Maximization (EM) algorithm. Experiments in a texture classification task are conducted to evaluate this extended modeling on several color texture databases, namely popular Vistex, 167-Vistex and CUReT. These experiments show that our new mixture model competes with state-of-the-art on the experimented datasets.

This paper presents a new approach for unsupervised band selection in the context of hyperspectral imaging. The hyperspectral band selection (HBS) task is considered as a clustering problem: bands are clustered in the image space; one representative image is then kept for each cluster, to be part of the set of selected bands. The proposed clustering method falls into the family of information-maximization clustering, where mutual information between data features and cluster assignments is maximized. Inspired by a clustering method of this family, we adapt it to the HBS problem and extend it to the case of multiple image features. A pixel selection step is also integrated to reduce the spatial support of the feature vectors, thus mitigating the curse of dimensionality. Experiments with different standard data sets show that the bands selected with our algorithm lead to higher classification performance, in comparison with other state-of-the-art HBS methods.

In this paper, we aim at super-resolving a low-resolution texture under the assumption that a high-resolution patch of the texture is available. To do so, we propose a variational method that combines two approaches that are texture synthesis and image reconstruction. The resulting objective function holds a nonconvex energy that involves a quadratic distance to the low-resolution image, a histogram-based distance to the high-resolution patch, and a nonlocal regularization that links the missing pixels with the patch pixels. As for the histogram-based measure, we use a sum of Wasserstein distances between the histograms of some linear transformations of the textures. The resulting optimization problem is efficiently solved with a primal-dual proximal method. Experiments show that our method leads to a significant improvement, both visually and numerically, with respect to the state-of-the-art algorithms for solving similar problems.

Data, which lie in the space Pm, of m × m symmetric positive definite matrices, (sometimes called tensor data), play a fundamental role in applications, including medical imaging, computer vision, and radar signal processing. An open challenge, for these applications, is to find a class of probability distributions, which is able to capture the statistical properties of data in Pm, as they arise in real-world situations. The present paper meets this challenge by introducing Riemannian Gaussian distributions on Pm. Distributions of this kind were first considered by Pennec in 2006. However, the present paper gives an exact expression of their probability density function for the first time in existing literature. This leads to two original contributions. First, a detailed study of statistical inference for Riemannian Gaussian distributions, uncovering the connection between the maximum likelihood estimation and the concept of Riemannian centre of mass, widely used in applications. Second, the derivation and the implementation of an expectation-maximisation algorithm, for the estimation of mixtures of Riemannian Gaussian distributions. The paper applies this new algorithm, to the classification of data in Pm, (concretely, to the problem of texture classification, in computer vision), showing that it yields significantly better performance, in comparison to recent approaches.

Conventional feature extraction methods cannot fully exploit both the spectral and spatial information of hyperspectral imagery. In this paper, we propose an ensemble method of subspace independent component analysis (ICA) and edge-preserving filtering (EPF) for the classification of hyper-spectral data to achieve this task. First, several subsets are randomly selected from the original feature space. Second, ICA is used to extract spectral independent components followed by a recent and effective EPF method, rolling guidance filter (RGF), to produce spatial features. The spatial features are treated as the input of a random forest (RF) classifier. Finally, the classification results from each subset are integrated together to produce the final map. Experimental results on real hyperspectral data demonstrate the effectiveness of the proposed method. A sensitivity analysis of this new classifier is also performed.

This paper introduces a generalization of the Fisher vectors to the Riemannian manifold. The proposed descriptors, called Riemannian Fisher vectors, are defined first, based on the mixture model of Riemannian Gaussian distributions. Next, their expressions are derived and they are applied in the context of texture image classification. The results are compared to those given by the recently proposed algorithms, bag of Riemannian words and R-VLAD. In addition, the most discriminant Riemannian Fisher vectors are identified.

Images are usually represented by different groups of features, such as color, shape and texture attributes. In this paper, we propose a classification approach that integrates multiple features, such as spectral and spatial information. We refer this approach to multiple feature learning via rotation (MFL-R) strategy, which adopt a rotation-based ensemble method by using a data transformation approach. Five data transformation methods, including principal component analysis (PCA), neighborhood preserving embedding (NPE), linear local tangent space alignment (LLTSA), linearity preserving projection (LPP) and multiple feature combination via manifold learning and patch alignment (MLPA) are used in the MFL-R framework. Experimental results over two hyperspectral remote sensing images demonstrate that MFL-R with MLPA gains better performances and is not sensitive to the tuning parameters.

Rotation Forest (RoF) is a decision tree ensemble classifier, which uses random feature selection and data transformation techniques to improve both the diversity and accuracy of base classifiers. Traditional RoF only considers data transformation on spectral information. In order to further improve the performance of RoF, we introduce spectral-spatial data transformation into RoF and thus propose a spectral-spatial Rotation Forest (SSRoF). The proposed method is experimentally investigated on a hyperspectral remote sensing image collected by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) sensor. Experimental results indicate that the proposed methodology achieves excellent performance.

To obtain accurate classification results of hyperspectral images, both spectral and spatial information should be fully exploited in the classification process. In this paper, we propose a novel method using independent component analysis (ICA) and edge-preserving filtering (EPF) via an ensemble strategy for the classification of hyperspectral data. First, several subsets are randomly selected from the original feature space. Second, ICA is used to extract spectrally independent components followed by an effective EPF method, to produce spatial features. Two strategies (i.e., parallel and concatenated) are presented to include the spatial features in the analysis. The spectral-spatial features are then classified with a random forest or a rotation forest classifier. Experimental results on two real hyperspectral data sets demonstrate the effectiveness of the proposed methods. A sensitivity analysis of the new classifiers is also performed.

We consider the problem of recovering a high-resolution image from a pair consisting of a complete low-resolution image and a high-resolution but incomplete one. We refer to this task as the image zoom completion problem. After discussing possible contexts in which this setting may arise, we introduce a nonlocal regularization strategy, giving full details concerning the numerical optimization of the corresponding energy and discussing its benefits and shortcomings. We also derive two total variation-based algorithms and evaluate the performance of the proposed methods on a set of natural and textured images. We compare the results and get with those obtained with two recent state-of-the-art single-image super-resolution algorithms.

Estimating the local orientations of a directional structure may prove to be complicated especially for complex geometries for which conventional methods, including those based on the gradient, are unsuitable. In this paper, we propose a recursive method of orientation estimation based on the reconstruction of hypersurfaces and the definition of a model of the structure in these so-called complex areas. The estimation proves to be more accurate when executed on the flattened data and driven to the initial referential. This process recursively leads to a more accurate orientation field estimation. When applied to synthetic data, the method reduces significantly the estimation bias. For real directional textures, the resulting orientation field coincides better with the perceptual orientation of the directional structures.

This paper presents a new Bayesian approach to texture classification, yielding enhanced performance in the presence of intraclass diversity. From a mathematical point of view, it specifies an original EM algorithm for mixture estimation on Riemannian manifolds, generalising existing, non probabilistic, clustering analysis methods. For texture classification, the chosen feature space is the Riemannian manifold known as the Poincaré half plane, here denoted H, (this is the set of univariate normal distributions, equipped with Rao's distance). Classes are modelled as finite mixtures of Riemannian priors, (Riemannian priors are probability distributions, recently introduced by the authors, which represent clusters of points in H). During the training phase of classification, the EM algorithm, proposed in this paper, computes maximum likelihood estimates of the parameters of these mixtures. The algorithm combines the structure of an EM algorithm for mixture estimation, with a Riemannian gradient descent, for computing weighted Riemannian centres of mass.

Multivariate generalized Gaussian distribution (MGGD) has been an attractive solution to many signal processing problems due to its simple yet flexible parametric form, which requires the estimation of only a few parameters, i.e., the scatter matrix and the shape parameter. Existing fixed-point (FP) algorithms provide an easy to implement method for estimating the scatter matrix, but are known to fail, giving highly inaccurate results, when the value of the shape parameter increases. Since many applications require flexible estimation of the shape parameter, we propose a new FP algorithm, Riemannian averaged FP (RA-FP), which can effectively estimate the scatter matrix for any value of the shape parameter. We provide the mathematical justification of the convergence of the RA-FP algorithm based on the Riemannian geometry of the space of symmetric positive definite matrices. We also show using numerical simulations that the RA-FP algorithm is invariant to the initialization of the scatter matrix and provides significantly improved performance over existing FP and method-of-moments (MoM) algorithms for the estimation of the scatter matrix.

This paper introduces an intrinsic prior distribution for supervised classification of texture images. First, we introduce the intrinsic prior distribution as the normal law on a Riemannian manifold. Next, based on this definition, we derive the estimation and classification schemes. Finally, we propose an application for the classification of texture images. Experiments on the VisTex texture database are conducted and demonstrate the interest of the proposed intrinsic classification algorithm.

In this paper, we are interested in estimating global motions (homographies, 3D rotations, 3D Euclidean motions, etc.) as well as the covariance of the estimation errors from relative measurements by exploiting the Lie group structure of the motions. We propose a generative model based on the formulation of a concentrated Gaussian distribution on matrix Lie groups. In this context, the global motion estimation problem reduces to the minimization of the sum of squared intrinsic invariant (w.r.t the right action of the Lie group on itself) errors. We derive an iterated extended Kalman filter on matrix Lie groups from the Gauss-Newton formalism on matrix Lie groups, which exhibits a low computational complexity. Experimental results on simulated data, in the context of a consistent pose registration problem, show that the proposed algorithm significantly outperforms the state of the art approaches.

The classical super-resolution (SR) setting starts with a set of low-resolution (LR) images related by subpixel shifts and tries to reconstruct a single high-resolution (HR) image. In some cases, partial observations about the HR image are also available. Trying to complete the missing HR data without any reference to LR ones is an inpainting (or completion) problem. In this paper, we consider the problem of recovering a single HR image from a pair consisting of a complete LR and incomplete HR image pair. This setting arises in particular when one wants to fuse image data captured at two different resolutions. We propose an efficient algorithm that allows to take advantage of both image data by first learning nonlocal interactions from an interpolated version of the LR image using patches. Those interactions are then used by a convex energy function whose minimization yields a super-resolved complete image.

In this paper, we address the problem of rotation invariance in the context of texture retrieval. For this, we propose a framework based on the well-known copula theory which is considered one of the most powerful statistical tools. Prior to apply a such model, we first use the steerable pyramid SP as one of the most relevant transforms. Then, we build a steerable Gaussian copula model which offers a good fitting of the SP coefficients distribution while taking into consideration their rotation invariance property. Finally, we derive a closed-form of the Jefferey divergence as a similarity measure. The latter consists on an angular alignment between the query and the target texture features. Experiments have been conducted on USC database, good performances in term of retrieval rates are achieved compared to previously proposed copula models.

This letter addresses an estimation problem based on blurred and noisy observations of textured images. The goal is jointly estimating the 1) image model parameters, 2) parametric point spread function (semi-blind deconvolution) and 3) signal and noise levels. It is an intricate problem due to the data model non-linearity w.r.t. these parameters. We resort to an optimal estimation strategy based on Mean Square Error, yielding the best (non-linear) estimate, namely the Posterior Mean. It is numerically computed using a Monte Carlo Markov Chain algorithm: Gibbs loop including a Random Walk Metropolis-Hastings sampler. The novelty is double: i) addressing this fully parametric threefold problem never tackled before through an optimal strategy and ii) providing a theoretical Fisher information-based analysis to anticipate estimation accuracy and compare with numerical results.

In the framework of texture image retrieval, a new family of stochastic multivariate modeling is proposed based on Gaussian Copula and wavelet decompositions. We take advantage of the copula paradigm, which makes it possible to separate dependence structure from marginal behavior. We introduce two new multivariate models using, respectively, generalized Gaussian and Weibull densities. These models capture both the subband marginal distributions and the correlation between wavelet coefficients. We derive, as a similarity measure, a closed form expression of the Jeffrey divergence between Gaussian copula-based multivariate models. Experimental results on well-known databases show significant improvements in retrieval rates using the proposed method compared with the best known state-of-the-art approaches.

Natural texture images exhibit a high intra-class diversity due to different acquisition conditions (scene enlightenment, perspective angle, ... ). To handle with the diversity, a new supervised classification algorithm based on a parametric formalism is introduced: the K-centroids-based classifier (K-CB). A comparative study between various supervised classification algorithms on the VisTex and Brodatz image databases is conducted and reveals that the proposed K-CB classifier obtains relatively good classification accuracy with a low computational complexity.

This paper introduces a centroid-based (CB) supervised classification algorithm of textured images. In the context of scale/orientation decomposition, we demonstrate the possibility to develop centroid approach based on multivariate stochastic modeling. The main interest of the multivariate modeling comparatively to the univariate case is to consider spatial dependency as additional features for characterizing texture content. The aim of this paper is twofold. Firstly, we introduce the Spherically Invariant Random Vector (SIRV) representation for the modeling of wavelet coefficients. Secondly, from the specific properties of the SIRV process, i.e. the independence between the two sub-processes of the compound model, we derive centroid estimation scheme. Experiments from various conventional texture databases are conducted and demonstrate the interest of the proposed classification algorithm.

The paper provides a spatio-temporal change detection framework for the analysis of image time series. In this framework, the detection of changes in time is addressed at the image level by using a matrix of cross-dissimilarities computed upon wavelet and curvelet image features. This makes possible identifying the acquisitions of interest: the acquisitions that exhibit singular behavior with respect to their neighborhood in the time series, and those that are representatives of some stationary behavior. These acquisitions of interest are compared at the pixel level to detect spatial changes characterizing the evolution of the time series. Experiments carried out over European Remote Sensing (ERS) and TerraSAR-X time series highlight the relevancy of the approach for analyzing synthetic aperture radar image time series.

This brief derives a 2-D spectrum estimator from some recent results on the statistical properties of wavelet packet coefficients of random processes. It provides an analysis of the bias of this estimator with respect to the wavelet order. This brief also discusses the performance of this wavelet-based estimator, in comparison with the conventional 2-D Fourier-based spectrum estimator on texture analysis and content-based image retrieval. It highlights the effectiveness of the wavelet-based spectrum estimation.

This paper studies the performance of the maximum likelihood estimators (MLE) for the parameters of multivariate generalized Gaussian distributions. When the shape parameter belongs to ]0, 1[, we have proved that the scatter matrix MLE exists and is unique up to a scalar factor. After providing some elements about this proof, an estimation algorithm based on a Newton-Raphson recursion is investigated. Some experiments illustrate the convergence speed of this algorithm. The bias and consistency of the scatter matrix estimator are then studied for different values of the shape parameter. The performance of the shape parameter estimator is finally addressed by comparing its variance to the Cramér-Rao bound.

This paper presents a new multivariate elliptical distribution, namely the multivariate generalized Gamma times an Uniform (MGΓU) distribution. Because it generalizes the multivariate generalized Gaussian distribution (MGGD), the MGΓU distribution is able to fit a wider range of signals. For the bivariate case, we provide a closed-form of the KullbackLeibler divergence (KLD). We propose the MGΓU distribution for modeling chrominance wavelet coefficients and exercise it in a texture retrieval experiment. A comparative study between some multivariate models on the VisTex and Outex image database is conducted and reveals that the use of the MGΓU distribution of chromiance wavelet coefficient allows an indexing gain compared to other classical approaches such as MGGD and Copula based model).

The paper addresses the analysis and interpretation of second order random processes by using the wavelet packet transform. It is shown that statistical properties of the wavelet packet coefficients are specific to the filtering sequences characterizing wavelet packet paths. These statistical properties also depend on the wavelet order and the form of the cumulants of the input random process. The analysis performed points out the wavelet packet paths for which stationarization, decorrelation and higher order dependency reduction are effective among the coefficients associated with these paths. This analysis also highlights the presence of singular wavelet packet paths: the paths such that stationarization does not occur and those for which dependency reduction is not expected through successive decompositions. The focus of the paper is on understanding the role played by the parameters that govern stationarization and dependency reduction in the wavelet packet domain. This is addressed with respect to semi-analytical cumulant expansions for modeling different types of nonstatonarity and correlation structures. The characterization obtained eases the interpretation of random signals and time series with respect to the statistical properties of their coefficients on the different wavelet packet paths.

The paper addresses content-based image retrieval from texture data bases, by using stochastic modeling in the wavelet domain. It pro poses an analysis of the key parameters involved in such a content based texture retrieval. These parameters are the wavelet order and the goodness-of-fit measure used to select the best family of distributions for modeling the subband wavelet coefficients. It is shown that taking suitable parameters into consideration makes it possible to attain high retrieval rates in content-based texture retrieval.

This paper presents a new wavelet based retrieval approach based on Spherically Invariant Random Vector (SIRV) modeling of wavelet subbands. Under this multivariate model, wavelet coefficients are considered as a realization of a random vector which is a product of the square root of a scalar random variable (called multiplier) with an independent Gaussian vector. We propose to work on the joint distribution of the scalar multiplier and the multivariate Gaussian process. For measuring similarity between two texture images, the geodesic distance is provided for various multiplier priors. A comparative study between the proposed method and conventional models on the VisTex image database is conducted and indicates that SIRV modeling combined with geodesic distance achieves higher recognition rates than classical approaches.

We have studied two efficient sampling methods, Langevin and Hessian adapted Metropolis Hastings (MH), applied to a parameter estimation problem of the mathematical model (Lorentzian, Laplacian, Gaussian) that describes the Power Spectral Density (PSD) of a texture. The novelty brought by this paper consists in the exploration of textured images modeled by centered, stationary Gaussian fields using directional stochastic sampling methods. Our main contribution is the study of the behavior of the previously mentioned two samplers and the improvement of the Hessian MH method by using the Fisher information matrix instead of the Hessian to increase the stability of the algorithm and the computational speed. The directional methods yield superior performances as compared to the more popular Independent and standard Random Walk MH for the PSD described by the three models, but can easily be adapted to any target law respecting the differentiability constraint. The Fisher MH produces the best results as it combines the advantages of the Hessian, i.e., approaches the most probable regions of the target in a single iteration, and of the Langevin MH, as it requires only first order derivative computations.

This paper presents a new texture retrieval algorithm based on elliptical distributions for the modeling of wavelet sub- bands. For measuring similarity between two texture images, the geodesic distance (GD) is considered. A closed form for fixed shape parameters and an approximation when assuming the geodesic coordinate functions as straight lines are given. Taken into various elliptical choices, the multivariate Laplace and G0distributions are introduced for modeling respectively the color cue and spatial dependencies of the wavelet coefficients. A multi-model classification approach is then proposed to combine the similarity measures. A comparative study between some multivariate models on the VisTex image database is conducted and reveals that the combination of the multivariate Laplace modeling for the color dependency and the multivariate G0modeling for spatial one achieves higher recognition rates than other approaches.

In this paper we address the problem of location recognition from visual lifelogs by leveraging visual features and temporal information in an unified framework. The proposed method features a co-training approach that takes advantage of both labeled and unlabeled data using a confidence measure we propose for this task. It exploits jointly two SVM classifiers on two types of visual features as well as the temporal continuity of the video through temporal accumulation scheme. We demonstrate experimentally on the publicly available IDOL2 dataset that the algorithm yields performance improvement due to its ability to exploit jointly multiple cues, time and unlabeled data.

In the framework of wavelet-based analysis, this paper deals with texture modeling for classification or retrieval systems using non-Gaussian multivariate statistical features. We propose a stochastic model based on Spherically Invariant Random Vectors (SIRVs) joint density function with Weibull assumption to characterize the dependences between wavelet coefficients. For measuring similarity between two texture images, the Kullback-Leibler divergence (KLD) between the corresponding joint distributions is provided. The evaluation of model performance is carried out in the framework of retrieval system in terms of recognition rate. A comparative study between the proposed model and conventional models such as univariate Generalized Gaussian distribution and Multivariate Bessel K forms (MBKF) is conducted.

In this paper, a new formulation based on bidirectional composition on Lie groups (BCL) for parametric gradient-based image alignment is presented. Contrary to the conventional approaches, the BCL method takes advantage of the gradients of both template and current image without combining them a priori. Based on this bidirectional formulation, two methods are proposed and their relationship with state-of-the-art gradient based approaches is fully discussed. The first one, i.e., the BCL method, relies on the compositional framework to provide the minimization of the compensated error with respect to an augmented parameter vector. The second one, the projected BCL (PBCL), corresponds to a close approximation of the BCL approach. A comparative study is carried out dealing with computational complexity, convergence rate and frequence of convergence. Numerical experiments using a conventional benchmark show the performance improvement especially for asymmetric levels of noise, which is also discussed from a theoretical point of view.

A new method for template-based image alignment is presented in this paper. A gradient-based optimization of motion compensated error difference is addressed to solve the image alignment problem. Its novelty lies in the minimized error, which considers the bi-directional compensation of the reference template and the current image onto each other in a constrained asymmetrical fashion. The proposed approach is shown to be a generalization of previous state-of-the art gradient-based methods. An experimental evaluation is provided to show how the new method outperforms the former in presence of noisy images, and to give some insights into its properties.

In this paper, we investigate both linear and circular stochastic models in the context of texture discrimination. These models aim at representing the magnitudes and orientations obtained by a complex wavelet decomposition, such as the steerable pyramid.The novelty consists in considering specific parametric models for circular data such as von Mises and psi- distributions to describe the distributions of orientations. Particular attention is paid to the choice of a metric and to its adequation to the models. Indexing experiments are conducted to quantitatively evaluate the performances of the proposed models and of the chosen matrices, i.e. the L1and Kullback-Leibler distances.

This paper deals with texture modeling for classification or retrieval systems using multivariate statistical features. The proposed features are defined by the hyperparameters of a copula-based multivariate distribution characterizing the coefficients provided by image decomposition in scale and orientation. As it belongs to the multivariate stochastic models, the copulas are useful to describe pairwise non-linear association in the case of multivariate non-Gaussian density. In this paper, we propose the d-variate Gaussian copula associated to univariate gamma densities for modeling the texture. Experiments were conducted on the VisTex database aiming to compare the recognition rates of the proposed model with the univariate generalized Gaussian model, the univariate Gamma model, and the generalized Gaussian copula-based multivariate model.

This paper deals with texture analysis based on multiscale stochastic modeling. In contrast to common approaches using symmetric marginal probability density functions of subband coefficients, experimental manipulations show that the symmetric shape assumption is violated for several texture classes. From this fact, we propose in this paper to exploit this shape property to improve texture characterization. We present Asymmetric Generalized Gaussian density as a model to represent detail subbands resulting from multiscale decomposition. A fast estimation method is presented and closed-form of Kullback-Leibler divergence is provided in order to validate the model into a retrieval scheme. The experimental results indicate that this model achieves higher recognition rates than the conventional approach of using the Generalized Gaussian model where asymmetry was not considered.

In this paper, we deal with the problem of feature extraction in content-based image retrieval (CBIR) using statistical approach. A Gussianization procedure based on parametric density assumptions of steerable pyramid coefficients is proposed. The extraction method of features including the Gaussianization step allows us to limit the order of the statistical model used to characterize the image textures. The performances of the proposed method are analyzed on a database of texture images and compared with the performances of other textures proposed in previous.

This paper presents a robust unsupervised framework for 3D seismic data flattening. The resulting volume, called GeoTime cube, brings to light history of sedimentary deposits which is a key issue in petroleum prospecting. The proposed method makes it possible to obtain the transformation by transcribing fundamental principles of geophysics in image processing. The first step is a sedimentary layer reconstruction, the second one consists in numbering them according to their relative geological age and the last one computes a transformation in order to clearly represent them in a flattened way. Finally, the results obtained by our method compared to an existing one show that many relevant information can be extracted from GeoTime cubes and the final flattened data enhances the seismic structures identification.

This paper deals with a 2-D high resolution frequency estimation method dedicated to corrupted image by outliers. Outliers are particular data points that do not obey the assumed model. In this framework the well-known model of the sum of complex exponentials fails for the smallest fraction of a data set, which causes the classical estimators to produce inaccurate results. To alleviate this drawback, we propose a new robust iterative Levenberg-Marquardt (LM)-based method. The three main steps of the method we propose are as follows. First, we define a weight function based on the influence function which allows one to detect and correct the "wrong" data. The influence function measures the influence of a datum on the value of the parameter estimate. It is inspired from the so-called M-estimator. Second, a 2-D extension of the large sample approximation of the Maximum likelihood (ML) estimator is developed in order to estimate the image parameters. Third, the Levenberg-Marquardt (LM) technique is used to ensure the convergence of the ML estimator by performing the detection of "wrong" data for each iteration. The effectiveness of the proposed method is illustrated by some numerical simulations.

In this paper a new video tracking process is proposed. Our approach uses two description levels for the segmentation mechanism. The first one, the pixel level, performs the extraction of every point in movement, forming, as result, connected regions. For this stage, a new adaptive reference image (ARI) algorithm is described, based on a derivative approach. This choice presents a robust method facing changes in illumination. The second level, the region level, is driven by the contents of a model set. This high level object description, defined by geometric attributes and a motion model lets us associate the ARI process with the handle objects. This stage sets this up as a Bayesian model formulation. The identification and updating process is done by a modified expectation-maximization (EM) algorithm. This step establishes the relationship between regions (the ARI process output) and the object models (OM). For each correspondence, the EM procedure is started, updating every attribute of the object description. This framework defines a complete unsupervised tracking procedure, robust regarding occlusions, over- or sub-segmentations and background brightness variations. Several real traffic examples are included at the end of the paper.

This article introduces the estimation of the optical flow, based on the combination of forward and backward constraint motion equations. Among all the existing techniques allowing the retrieval of the motion, many methods start from the assumption of recorded image brightness dependency after displacement. This dependency spans a derivative formulation, so called motion constraint equation, based on the spatial transformation. In this framework, for fixed temporal positions, the motion of a rigid object can be described by a forward and backward exploitation of the motion constraint equation, for the same spatial position. This setting in equation enables us to establish a new relation called a forward-backward optical flow equation.

Classical high resolution spectral analysis methods for the estimation of complex sinusoids parameters require the a priori knowledge of the number of sinusoids. In most situations, the correctness of this knowledge is crucial. In this paper, we present a new high resolution sub-space method. Its novelty stems from the fact that the analysis of complex sinusoids is considered as a joint "detection-estimation" issue. Simulation results and an application on real radar signals are presented to illustrate the efficiency of this method.

This paper addresses the problem of estimating the frequency content of a two-dimensional object, e.g. an image or a set of multi-sensor snapshots, stored in a matrix. The basic assumption is that the data matrix consists of a sum of 2-D complex sinusoids. Our algebraically coupled matrix pencils (ACMP) algorithm splits the 2-D problem into two related 1-D estimation problems. In each direction the frequencies are estimated using a computationally efficient ESPRIT-like subspace algorithm. A further increase in efficiency is due to the algebraic pairing of the horizontal and vertical estimates.>