The proposal of deep learning (DL) solutions for SAR image despeckling is recently widespread. Such solutions have been mainly designed in a DL perspective by leveraging the training and validation stage on the use of typical norm-based cost functions. For going beyond the DL perspective, in this paper we propose a SAR based validation stage by using SAR assessing metrics in the design and hyper-parameter selection of neural networks. In a first phase, SAR assessing metrics may be used only as validation metrics with the aim of highlighting critical issues that can not be spotted with standard image-processing quality metrics. In a second phase, the same SAR assessing metrics may be used directly for enhancing the DL solution by the addressing specific issues arisen during the previous SAR based validation stage. To this aim, three different DL SAR despeckling solutions and four different SAR assessing metrics have been considered. The outcome of this analysis shows the importance of including SAR knowledge in the training and validation stages of the design of a DL solution for SAR image despeckling.

In this letter, considering the effectiveness of 2-D principal component analysis (2DPCA) on the exploration of local spatial relationships, a reconstruction-based 2DPCA (Rec-2DPCA) operation was designed for feature extraction and injected into the architecture of PCANet for change detection of bitemporal synthetic aperture radar (SAR) image. Specifically, as the projection of an image patch on one eigenvector computed by 2DPCA breaks the one-to-one relationship between feature map and eigenvalue, we adopted Rec-2DPCA at various network layers and developed two variants of PCANet, namely, 2DPCANet and (2-D + 1-D)PCANet. In the experiments, using three real SAR image datasets, we analyzed the performance of all comparison methods, and our proposals achieved a more appealing performance than other methods.

Abstract Glioblastoma multiforme (GBM) is the most prevalent and aggressive primary brain tumour that has the worst prognosis in adults. Currently, the automatic segmentation of this kind of tumour is being intensively studied. Here, the automatic three‐dimensional segmentation of the GBM is achieved with its related subzones (active tumour, inner necrosis, and peripheral oedema). Preliminary segmentations were first defined based on the four basic magnetic resonance imaging modalities and classic image processing methods (multithreshold Otsu, Chan–Vese active contours, and morphological erosion). After an automatic gap‐filling post processing step, these preliminary segmentations were combined and corrected by a supervised artificial neural network of multilayer perceptron type with a hidden layer of 80 neurons, fed by 30 selected radiomic features of gray intensity and texture. Network classification has an overall accuracy of 83.9%, while the complete combined algorithm achieves average Dice similarity coefficients of 89.3%, 80.7%, 79.7%, and 66.4% for the entire region of interest, active tumour, oedema, and necrosis segmentations, respectively. These values are in the range of the best reported in the present bibliography, but even with better Hausdorff distances and lower computational costs. Results presented here evidence that it is possible to achieve the automatic segmentation of this kind of tumour by traditional radiomics. This has relevant clinical potential at the time of diagnosis, precision radiotherapy planning, or post‐treatment response evaluation.

Deep Learning methods require immense amounts of labeled data to provide reasonable results. In computer vision applications, and more specifically in despeckling SAR (Synthetic Aperture Radar) images, due to the speckle content, there is no ground truth available. To test the performances of despeckling filters, the common protocol is to synthetically corrupt optical images with a suitable speckle model and then, after filtering, well-known metrics are obtained. Then, filters are tested on actual SAR data. However, even the most elaborated speckle models are far from accounting for the complex mechanisms related to SAR images. In this paper, a methodology to design a realistic dataset is proposed. Actual SAR images of the same scene acquired with the same sensor on different dates, then they are properly co-registered and averaged to get a ground truth-like reference image to objectively evaluate the performance of a despeckling method. To show the benefits of the proposed methodology, a deep learning approach is used to filter the data by using the designed dataset, which will be called the “SAR model”. Then they are compared with the standard protocol by using synthetically corrupted optical images, which will be the “Synthetic model”. One last validation is performed by filtering the same images with FANS, a well-known despeckling filter and compared with the results obtained with autoencoder. The validation on actual SAR data not included in the training phase validates the proposed methodology. From the results shown, it is recommended to test filters on the proposed more realistic dataset.

Abstract Mobile communications systems are affected by what is known as fading, which is a well-known problem largely studied for decades. The direct consequence of fading is the complete loss of signal (or a large decrease of the received power). Rayleigh fading is a reasonable model for wireless channels although, Nakagami-m distribution seems better suited to fitting experimental data. In this paper we obtain the Nakagami-m distribution as a composite (mixture) of the Rayleigh distribution, a result which as far as we know it has not been shown in the literature. This representation of the Nakagami-m distribution facilitates computations of the average BER (Bit Error Rate) for DPSK (Differential Phase Shift Keying) and MSK (Minimum-Shift Keying) modulations for this distribution and higher moments of them, which is of great applicability to modeling wireless fading channels. Furthermore, a simple, not depending on any special function, apart of the Gamma function, bivariate version of the Nakagami-m distribution is also proposed as a special case of the multivariate version which is also presented. The proposed composite distribution is simulated through the standard procedure of summation of phasors, and results for the new closed-form measures for the MSK modulation are also shown. From that it is clear that the alternative formulation of the Nakagami-m distribution allows for easier modeling of fading fading-shadowing wireless channels through the new explicit second order statistics metrics. is well suited for modelling fading-shadowing wireless channels.

SAR (Synthetic Aperture Radar) sensors are fundamental tools for the Earth Observation. Actually, SAR images are affected by a multiplicative noise speckle that require a filtering step crucial for further application: classification, detection, etc. As in all image processing task, deep learning has been widely used for SAR image despeckling in the last years. Many methods have been proposed with different architectures, cost functions, training approaches, showing impressive performance. Actually, differently from natural domain denoising, an extensive comparison of such methods is still missing. As matter fact, such methods focus their comparison on few testing images. The aim of this paper is to propose and carry out the comparison among DL (Deep Learning) based methods not only in the testing phase but explointg the validation dataset used during the training for evaluating performance on SAR based metrics. The aim is to give an assessment on a more wide scenarios.

SAR imagery is always accompanied by speckle because it is obtained with coherent illumination. Given the multiplica-tive property of speckle, ratio-based metrics are widely used in SAR image quality assessment. In this paper, using the hy-potheses testing principle, a likelihood ratio test (LRT) based metric was designed for similarity measure in intensity SAR imagery. Then, using non-local means (NLM) scheme, an LRT-based NLM filter was proposed for speckle reduction of SAR imagery. In the experiments, using simulated intensity speckled images, the performance of the designed method is analyzed.

Speckle filtering is an important step for target detection in SAR images since this effect makes it difficult or even impossible to extract information from these images. There are several filters available in the literature although evaluating their performances is not a trivial task since it requires comparing the filtered images with a speckle-free image, which is generally unknown. This evaluation is even more complex when the features in the images are heterogeneous, for example, from tropical forests. The objective of this study is to evaluate the performance of the Lee, deGrandi, GammaMAP, single Anisotropic Nonlinear Diffusion (ANLD), multitemporal ANLD, Fast Adaptive Nonlocal SAR (FANS), and Fast GPU-Based Enhanced Wiener filters to reduce the speckle present in the COSMO-SkyMed Stripmap X-band images from the Brazilian Amazon forest region. The evaluation was conducted qualitatively through the visual inspection of the ratio image and the edge detection in the ratio images and quantitatively through the$\alpha \beta $estimator and other statistical parameters of the filtered images. The GammaMAP filter showed the best performances, both qualitatively and quantitatively, and the FANS filter only qualitative.

In this letter, we propose a new despeckling filter for fully polarimetric synthetic aperture radar (PolSAR) images defined by 3×3 complex Wishart distributions. We first generalize the well-known structure tensor to deal with PolSAR data which allows to efficiently measure the dominant direction and contrast of edges. The generalization includes stochastic distances defined in the space of the Wishart matrices. Then, we embed the formulation into an anisotropic diffusion-like schema to build a filter able to reduce speckle and preserve edges. We evaluate its performance through an innovative experimental setup that also includes Monte Carlo analysis. We compare the results with a state-of-the-art polarimetric filter.

The NLM (Nonlocal Means) approach is an effective framework for noise reduction. It relies on building, for each pixel, a convolution matrix whose entries are measures of similarities between patches. Such measures have to be computed between positive-definite Hermitian matrices when it comes to Polarimetric Synthetic Aperture Radar (PolSAR) imagery. Speckle reduction in this kind of images is a difficult task, as it is expected that several properties are well preserved by the filter. Torres et al. (2014) used a test statistic between Wishart distributions based on the Hellinger distance to compute such measures of similarity. In this work we assess the impact of using this and two other stochastic distances (Kullback-Leibler and Bhattacharya) under the same framework. The comparison is made using mean preservation, equivalent number of looks, edge correlation and the structural similarity index.

In this letter, we propose a supervised method for the classification of fully polarimetric synthetic aperture radar (PolSAR) images based on active contour models. We use an “a priori” estimation, obtained from training data, of the complex Wishart distributions of the different types of regions in the image (for instance, water, crops, grass, forest or urban). The information of the Wishart distributions is included in the active contour models to guide the level set evolution. We study the case of two classes and the case of three or more classes separately. We present some experimental results on the synthetic data and real PolSAR images to show the performance of the proposed model. The results are compared with other well-known supervised classification methods, and, for actual PolSAR data, our method shows an overall precision of 94.31% and a $\kappa $ coefficient of 0.937.

In this paper we study the local behavior of Fully PolSAR (Polarimetric Synthetic Aperture Radar) images defined by 3×3 complex Wishart distributions. We propose a generalization of the well-known structure tensor to Fully PolSAR images, as well as a measure of the smoothness of such predominant direction. This measure provides local information as the predominant direction of maximum variation of the complex Wishart distribution in a neighborhood of an image domain point ( x, y). We use stochastic distances defined in the space of Wishart matrices to generalize de structure tensor to PolSAR images. The study of the local behavior of PolSAR images address a number of processing and analysis problems. In particular, in this paper we apply this new approach to edge estimation using the magnitude of the PolSAR image variation provided by the generalized structure tensor. We also show promising results for simulated and actual Pol-SAR data.

Synthetic Aperture Radar (SAR) imaging play a central role in Remote Sensing applications due to, among other important features, its ability to provide high-resolution, day-and-night and almost weather-independent images. SAR images are affected from a granular contamination, speckle noise, that can be described by a multiplicative model. Many de-speckling techniques have been proposed in the literature, as well as measure of the quality of the results they provide. Speckle filters provide X, estimator of the true image X̂, based solely on the observed data Z, then an ideal estimator would be the one for which the ratio of the observed image to the filtered one Π = Z/X̂ is only speckle. The quality of the filter can be, then, assessed by its closeness to this hypothesis. We tackle the problem of quantitatively measuring the quality of speckle filters by the criterion of lack of structure in the ratio image they produce. We propose the use of Haralick's textural features for the identification of remaining structures in ratio images. In order to do so, we first analyze the distribution of such features under the null hypothesis (H0) of absence of structure; then we sample from ratio images with barely visible remaining structures.

In this paper, we study lens distortion for still images considering two well-known distortion models: the two-parameter polynomial model and the two-parameter division model. We study the invertibility of these models, and we mathematically characterize the conditions for the distortion parameters under which the distortion model defines a one-to-one transformation. This ensures that the inverse transformation is well defined and the distortion-free image can be properly computed, which provides robustness to the distortion models. A new automatic method to correct the radial distortion is proposed, and a comparative analysis for this method is extensively performed using the polynomial and the division models. With the aim of obtaining an accurate estimation of the model, we propose an optimization scheme which iteratively improves the parameters to achieve a better matching between the distorted lines and the edge points. The proposed method estimates two-parameter radial distortion models by detecting the longest distorted lines within the image. This is done by applying the Hough transform extended with a radial distortion parameter. Next, a two-parameter model is estimated using an iterative nonlinear optimization scheme. This scheme aims at minimizing the distance from the edge points to their associated lines by adjusting the two distortion parameters as well as the coordinates of the center of distortion. We present some experiments on real images with significant distortion to show the ability of the proposed approach to correct the radial distortion. A visual and quantitative comparison between both automatic two-parameter model estimations indicates that the division model is more efficient for those images showing strong distortion.

Speckle reduction is an important problem in synthetic aperture radar (SAR) image analysis. Recent years have seen how Bayesian filters emerge as the natural extension of the nonlocal means filters, providing a general framework to deal with multiplicative (speckle) noise. In this paper, we present an easy-to-use software tool applying an evolutionary algorithm to optimize a Bayesian nonlocal means filter with sigma preselection for denoising SAR images. The desired result is a filtered image having a significative reduction in its variance but preserving the original mean value of the noisy image. A mixed-integer constrained optimization problem is stated and solved with the human intervention, where the user assists the evolutionary algorithm to reduce the noisy image variance under the restriction of keeping the mean value of the noisy SAR image within a predetermined interval of acceptance. We apply the methodology to a set of synthetic and real SAR speckle corrupted images. The results through the evaluation of objective global and local quality criteria show the excellent potential of the proposal.

ABSTRACT The signal received in a mobile radio environment exhibits rapid signal level fluctuations which are generally Rayleigh‐distributed. These result from interference by multiple scattered radio paths between the base station and the mobile receptor. Fading‐shadowing effects in wireless channels are usually modelled by means of the Rayleigh–Lognormal distribution (RL), which has a complicated integral form. The K‐distribution (K) is similar to RL but it has a simpler form and its probability density function admits a closed form; however, due to the Bessel function, parameter estimates are not direct. Another possible approach is that of the Rayleigh‐inverse Gaussian distribution (RIG). In this paper, an alternative is presented, a generalisation of the Rayleigh distribution which is simpler than the RL, K and RIG distributions, and thus more suitable for the analysis and design of contemporary wireless communication systems. Closed‐form expressions for the bit error rate (BER) for differential phase‐shift keying (DPSK) and minimum shift keying (MSK) modulations with the proposed distribution are obtained. Theoretical results based on statistically well‐founded distance measurements validate the new distribution for the cases analysed. Copyright © 2011 John Wiley & Sons, Ltd.

We present an interactive easy-to-use software package, based on an evolutionary algorithm, to perform adaptive anisotropic diffusion speckle filtering for synthetic aperture radar (SAR) images. As a main difference from other methodologies, there is an integration of a SAR-image human expert who provides a subjective validation to complete the diffusion filter design. We have applied an interactive methodology to a set of SAR images, and we compared the results with those obtained by other speckle reduction filters. The results, through the evaluation of objective global and local quality criteria, show the potential of the proposal.