Denoising and demosaicking long-wave infrared (LWIR) division-of-focal-plane (DoFP) polarization images are crucial for various vision applications. However, existing methods rely on the sequential application of individual denoising and demosaicking processes, which may result in the accumulation of errors produced by each process. To address this issue, we propose a joint denoising and demosaicking method for LWIR DoFP images based on a three-stage progressive deep convolutional neural network. To ensure the generalization ability of this network, it is essential to have adequate training data that closely resembles real data. Therefore, we model the complex noise sources that affect LWIR DoFP images as mixed Poisson-Additive-Stripe noise and construct a least-squares problem based on the polarization measurement redundancy error to estimate the parameters of this model on real images. Subsequently, the estimated noise parameters are used to generate training data that enables the network to learn accurate polarization image statistics and improve its generalization ability. The experimental results demonstrate the effectiveness of the proposed method in enhancing the image restoration performance on real LWIR DoFP polarization data.

Assessing the quality of polarization images is of significance for recovering reliable polarization information. Widely used quality assessment methods including peak signal-to-noise ratio and structural similarity index require reference data that is usually not available in practice. We introduce a simple and effective physics-based quality assessment method for polarization images that does not require any reference. This metric, based on the self-consistency of redundant linear polarization measurements, can thus be used to evaluate the quality of polarization images degraded by noise, misalignment, or demosaicking errors even in the absence of ground-truth. Based on this new metric, we propose a novel processing algorithm that significantly improves demosaicking of division-of-focal-plane polarization images by enabling efficient fusion between demosaicking algorithms and edge-preserving image filtering. Experimental results obtained on public databases and homemade polarization images show the effectiveness of the proposed method.

We analyze the application of Shannon entropy and Bhattacharyya distance for the characterization of polarimetric interferometric synthetic aperture radar (PolInSAR) images. We discuss the relevance of the decomposition of the Shannon entropy of a homogeneous region into the sum of three physically meaningful terms that depend respectively on intensity, polarimetry, and interferometry. We also discuss an analogous decomposition property for the Bhattacharyya distance, which is an efficient measure of the contrast between two regions. We illustrate the application of this approach for analyzing different contributions in the information content of PolInSAR images.

We propose a new approach for evaluating the contribution of the different channels of polarimetric and interferometric synthetic aperture radar (PolInSAR) images. For that purpose, we demonstrate that the Bhattacharyya distance between the probability density functions of neighboring regions in the image provides an efficient scalar contrast measure. We show that the analysis of this contrast measure allows one to precisely characterize the contribution of each channel for different system configurations, including intensity, polarimetric, and interferometric images. We illustrate this approach using a real synthetic aperture radar image to compare several polarimetric system architectures. Since PolInSAR imaging configurations can correspond to complex and expensive systems, the proposed method can be helpful in system imaging optimization.

In this paper, we show that polarimetric and interferometric SAR (PolInSAR) images can be efficiently partitioned into homogeneous regions with a statistical technique based on minimization of a parameter-free criterion. This technique consists of finding a polygonal partition of the image that minimizes the stochastic complexity, assuming that the image is made of a tessellation of statistically homogeneous regions. The obtained results demonstrate that a global partition in statistically homogeneous regions of PolInSAR images can provide better results than a partition based on a single characteristic such as polarimetry or interferometry only.

We consider polarimetric images formed with coherent waves, such as in laser-illuminated imagery or synthetic aperture radar. A definition of the contrast between regions with different polarimetric properties in such images is proposed, and it is shown that the performances of maximum likelihood-based detection and segmentation algorithms are bijective functions of this contrast parameter. This makes it possible to characterize the performance of such algorithms by simply specifying the value of the contrast parameter.

We analyze level set implementation of region snakes based on the maximum likelihood method for different noise models that belong to the exponential family. We show that this approach can improve segmentation results in noisy images and we demonstrate that the regularization term can be efficiently determined using an information theory-based approach, i.e., the minimum description length principle. The criterion to be optimized has no free parameter to be tuned by the user and the obtained segmentation technique is adapted to nonsimply connected objects.

We describe a general framework to design optimal image processing algorithms for polarimetric images formed with coherent radiations, which can be optical or microwave. Starting from the classical speckle model for coherent signals, we show that a wide class of algorithms to perform such tasks as detection, localization and segmentation depend on a simple statistics, which is the determinant of the coherency matrix estimated on a given region of the image. We use this property to design computationally efficient techniques for target/edge detection and image segmentation using statistical active contours and the minimum description length principle.

We address the problem of target segmentation and detection in coherent active polarimetric images. We consider polarimetric imagers which illuminate the scene with a laser beam and which measure how this light is depolarized by the scene. Since they are formed with coherent light, these "depolarization" images are perturbed by strong speckle-like noise, and by the spatial nonuniformity of the illumination beam. We design image processing algorithms which are adapted to the original statistical properties of these images. We address object segmentation using statistical active contours and target detection using maximum-likelihood (ML) algorithms. We show that the proposed algorithms perform well on both simulated and real-world images.

The optimal detection and location of a target in a scene in presence of non-additive noise is studied. The optimal processors in the maximum likelihood sense are determined for different white or correlated noise statistics. It is shown that the required computations are mainly correlations, and that there is no need of iterative methods.

In this paper we investigate the performance of a technique for face recognition based on the computation of 25 local autocorrelation coefficients. We use a large database of 11,600 frontal facial images of 116 persons, organized in training and test sets, for evaluation. Autocorrelation coefficients are computationally inexpensive, inherently shift-invariant and quite robust against changes in facial expression. We focus on the difficult problem of recognizing a large number of known human faces while rejecting other, unknown faces which lie quite close in pattern space. A multiresolution system achieves a recognition rate of 95%, while falsely accepting only 1.5% of unknown faces. It operates at a speed of about one face per second. Without rejection of unknown faces, we obtain a peak recognition rate of 99.9%. The good performance indicates that local autocorrelation coefficients have a surprisingly high information content.