Amrane Houacine

Redundant versus orthogonal wavelet decomposition for multisensor image fusion

Abstract The wavelet decomposition has become an attractive tool for fusing multisensor images. Usually, the input images are decomposed with an orthogonal wavelet in order to extract features, which are combined through an appropriate fusion rule. The fused image is then reconstructed by applying the inverse wavelet transform. In this paper, we investigate the use of the nonorthogonal (or redundant) wavelet decomposition as an alternative approach for feature extraction. By using test and remote sensing images, various fusion rules are considered and the detailed comparison indicates the superiority of this approach compared to the standard orthogonal wavelet decomposition for image fusion.

Statistical characterisation and modelling of SAR images

Statistical characterisation and modelling of SAR images is of great importance for developing classification algorithms and specialised filters for speckle noise reduction, among other applications. We present here the methods that estimate from the observed data the models that describe their statistical behaviour in a good way. Using the K distribution, the derived models depend on only one parameter whose estimation can be improved by using the bootstrap sampling method coupled with the Monte Carlo technique. An adequate representation of such models in the Pearson system allows physical interpretations. We show also that the K distribution-based models can be deduced through the use of Mellin multiplicative convolution, which has advantage in leading to an easier derivation. To confirm the judicious choice of the K distribution-based models, we provide a comparison with three other models that are often used in the literature.

Regularized fast recursive least squares algorithms for adaptive filtering

Fast recursive least squares (FRLS) algorithms are developed by using factorization techniques which represent an alternative way to the geometrical projections approach or the matrix-partitioning-based derivations. The estimation problem is formulated within a regularization approach, and priors are used to achieve a regularized solution which presents better numerical stability properties than the conventional least squares one. The numerical complexity of the presented algorithms is explicitly related to the displacement rank of the a priori covariance matrix of the solution. It then varies between O(5m) and that of the slow RLS algorithms to update the Kalman gain vector, m being the order of the solution. An important advantage of the algorithms is that they admit a unified formulation such that the same equations may equally treat the prewindowed and the covariance cases independently from the used priors. The difference lies only in the involved numerical complexity, which is modified through a change of the dimensions of the intervening variables. Simulation results are given to illustrate the performances of these algorithms. >

Principal component analysis of multispectral images using neural network

The conventional approach of PCA applied to multispectral images involves the computation of the spectral image covariance matrix and application of diagonalization procedures for extracting the eigenvalues and corresponding eigenvectors. When the number of spectral images grows significantly, the matrix computation and manipulation become practically inefficient and inaccurate due to round-off errors. These deficiencies make the conventional scheme inefficient for this application. We propose a neural network model that performs the PCA directly from the original spectral images without any additional non-neuronal computations or preliminary matrix estimation. The design of the network topology and input/output representation as well as the design of learning algorithms are carefully established. The convergence of the model is studied. Its application has been realized on real multispectral images. The obtained results show that the model performs well.

Regularized fast recursive least squares algorithms for finite memory filtering

Novel fast recursive least squares algorithms are developed for finite memory filtering, by using a sliding data window. These algorithms allow the use of statistical priors about the solution, and they maintain a balance between a priori and data information. They are well suited for computing a regularized solution, which has better numerical stability properties than the conventional least squares solution. The algorithms have a general matrix formulation, such that the same equations are suitable for the prewindowed as well as the covariance case, regardless of the a priori information used. Only the initialization step and the numerical complexity change through the dimensions of the intervening matrix variables. The lower bound of O(16m) is achieved in the prewindowed case when the estimated coefficients are assumed to be uncorrelated, m being the order of the estimated model. It is shown that a saving of 2m multiplications per recursion can always be obtained. The lower bound of the resulting numerical complexity becomes O(14m), but then the general matrix formulation is lost. >

Compound PCA-ICA neural network model for enhancement and feature extraction of multi-frequency polarimetric SAR imagery

Through its demixing operation, the potential use of independent components analysis (ICA) for multi-frequency polarimetric SAR imagery enhancement and feature extraction is demonstrated. A compound PCA-ICA neural network model is proposed, which consists of two levels of processing. The first one is the simultaneous diagonalization of the signal and signal-dependent noise covariance matrices using PCA transforms. The goal is to provide the PC images that are decorrelated and in which the SNR is improved. The second one consists of separating the noise from these images by providing new IC images in which the speckle is reduced. These images approach the PC ones and may be different only in their order and contrast. As a quantitative criterion, the contrast ratio is used, which value is smaller when the speckle is reduced. The model has been applied to the SIR-C data. The extracted features are quite effective for scene interpretation.

Accelerometer and Camera-Based Strategy for Improved Human Fall Detection

In this paper, we address the problem of detecting human falls using anomaly detection. Detection and classification of falls are based on accelerometric data and variations in human silhouette shape. First, we use the exponentially weighted moving average (EWMA) monitoring scheme to detect a potential fall in the accelerometric data. We used an EWMA to identify features that correspond with a particular type of fall allowing us to classify falls. Only features corresponding with detected falls were used in the classification phase. A benefit of using a subset of the original data to design classification models minimizes training time and simplifies models. Based on features corresponding to detected falls, we used the support vector machine (SVM) algorithm to distinguish between true falls and fall-like events. We apply this strategy to the publicly available fall detection databases from the university of Rzeszow’s. Results indicated that our strategy accurately detected and classified fall events, suggesting its potential application to early alert mechanisms in the event of fall situations and its capability for classification of detected falls. Comparison of the classification results using the EWMA-based SVM classifier method with those achieved using three commonly used machine learning classifiers, neural network, K-nearest neighbor and naïve Bayes, proved our model superior.

Model for multispectral and panchromatic image fusion

The Intensity Hue Saturation (IHS) transform is a widely used method to enhance the spatial resolution of multispectral images by substituting the Intensity component by the high resolution of the panchromatic image. However, such a direct substitution introduces important modifications on spectral properties. A more rigorous approach should consist in enhancing the spatial resolution of the intensity component through an appropriate combination with the panchromatic image. Such a combination is performed in the redundant wavelet domain by using a fusion model. SPOT images are used to illustrate the superiority of our approach compared to the IHS method for preserving spectral properties.

Fusion of multispectral and radar images in the redundant wavelet domain

We propose in this paper an integration method of the radar information in multispectral images without disturbing the spectral content. The main problem is to define a fusion rule that allows to take into account the characteristics of these images. Also, the main purpose of this paper lies in defining a new fusion rule performed in the redundant wavelet domain. This rule is based on the Mahalanobis distance applied on the wavelet coefficients. Instead of comparing coefficient-to- coefficient, the distance-to-distance comparison is performed. In this case the selected coefficient in the fused image will be the one that presents the large distance. This approach is applied to fusing the infrared band of SPOT with, respectively, RADARSAT and ERS images. The results show that spectral information is well preserved and there is a better information on the texture and the area roughness.

Automatic Classification of Human Body Postures Based on Curvelet Transform

This paper presents the design and implementation of a posture classification method. A new feature extraction strategy according to curvelet transform is provided for identifying the posture in images. First of all, human body is segmented. For this purpose, a background subtraction technique is applied. Then, a curvelet transform is used for extracting features from the posture image. To address the rotation invariance problem, five ratios are evaluated from the human body and they are also included in the set of features. Finally the human body postures are classified through support vector machines (SVM). Experimental results are obtained on the “Fall Detection” dataset. For evaluation, different state of the art statistical measures have been considered such as overall accuracy, the kappa coefficient, the F-measure coefficient, and the area under ROC curve (AUC) value. All of these evaluation measures demonstrate that the proposed approach provides a significant recognition rate.