Olivier Alata

Unsupervised segmentation for automatic detection of brain tumors in MRI

In this paper, we present a new automatic segmentation method for magnetic resonance images. The aim of this segmentation is to divide the brain into homogeneous regions and to detect the presence of tumors. Our method is divided into two parts. First, we make a pre-segmentation to extract the brain from the head. Then, a second segmentation is done inside the brain. Several techniques are combined like anisotropic filtering or stochastic model-based segmentation during the two processes. The paper describes the main features of the method, and gives some segmentation results.

Choice of a 2-D causal autoregressive texture model using information criteria

In the context of parametric modeling for image processing, we derive an estimation method for both the order and the parameters of 2-D causal autoregressive model with different geometries of support. Model parameters are estimated from a lattice representation, i.e. based on reflection coefficients. Lattice parameter estimation algorithms offer advantages compared to the Yule-Walker method: they do not require matrix inversion and their computation are robust and fast. For order selection, information criterion (IC) methods are the most commonly used. Therefore our order selection method is based on the combination of an IC and the prediction errors of models computed from the lattice parameter estimation algorithm. In this paper, we favour two consistent criteria compared to the nonconsistent Akaike criterion: the first criterion is a 2-D extension of Bayesian information criterion; the second criterion, noted φβ, extended here to the 2-D case, is a generalization drawn on Rissanens works. Simulations are provided on synthetic and natural textures with quarter plane support and non-symmetrical half plane support. We validate our results on natural textures using the Kullback divergence. The results show the interest of the combination of 2-DFLRLS algorithm and φβ, criterion to characterize natural textures.

Extension of the Schur-Cohn stability test for 2-D AR quarter-plane model

The Schur-Cohn test plays an essential role in checking the stability of one-dimensional (1D) random processes such as autoregressive (AR) models, via the so-called reflection coefficients, partial correlations, or Schur-Szego coefficients. In the context of two-dimensional (2D) random field modeling, one of the authors recently proposed a 2D AR quarter-plane model representation using 2D reflection coefficients estimated by a fast recursive adaptive algorithm. Based on such 2D reflection coefficients, we can therefore derive two necessary stability conditions for a 2D AR quarter-plane model. One of these conditions can be considered as an extension of the Schur-Cohn stability criterion based on the 2D reflection coefficients.

Transform image coding with global thresholding: application to baseline JPEG

Many image compression schemes perform the discrete cosine transform (DCT) to represent an image in frequency space. An analysis of a suite of images confirms that the luminance AC coefficients can be modeled by a Laplacian distribution. The distribution model can be used to drop the insignificant coefficients. In this paper we develop an image-adaptive JPEG encoding algorithm that incorporates global thresholding and near optimal quantization approach based on Lagrangian multiplier. Simulation results demonstrate that, with our thresholding technique, we can improve the reconstructed image quality compared to the one provided by other DCT image coding schemes without thresholding.

A new 2-D spectrum estimate using multichannel AR approach of 2-D fast RLS algorithms

In the framework of high resolution 2-D spectrum analysis, a new multichannel approach called harmonic mean horizontal vertical (HMHV) is proposed. It is based on 2-D fast recursive least squares (2-D FRLS) algorithms and their use for the computation of causal 2-D autoregressive (AR) parameters. This HMHV spectrum presents the following three main advantages on the 2-D spectrum estimated by the harmonic mean (HM) of the 2-D AR first and second quarter plane supports (QP1 and QP2) spectrum estimates: first, it presents the same biases and variances of estimation for the horizontal and vertical frequency components and improves in many cases the variances obtained with the HM method. Secondly, the single peak area (SPA) of the HMHV estimate is quite circular although the HM one looks like a skewed square indicating the existence of a best direction for the separation of two sinusoids. Thirdly, the new estimate presents less spurious peaks. This paper sums up the calculation of the different spectrum estimates and the experiments which lead to the conclusions.

2-D high resolution spectral estimation based on multiple regions of support

This paper deals with frequency estimation in the 2-D case when one has only few data points. We propose a method to estimate the frequencies of a sum of exponentials. This method is based on an original set of 2-D linear prediction models with new regions of support derived from the standard quarter plane support region. These models define various spectra which are finally combined by computing their harmonic mean. This method benefits from the subspace decomposition of the covariance matrix to perform well. It is demonstrated that the new regions of support improve the spectrum geometry and the estimation accuracy compared to the classical quarter plane (QP) support regions.

Parametric models of linear prediction error distribution for color texture and satellite image segmentation

In this article we present a Bayesian color texture segmentation framework based on the multichannel linear prediction error. Two-dimensional causal and non-causal real (in RGB color space) and complex (in IHLS and L^*a^*b^* color spaces) multichannel linear prediction models are used to characterize the spatial structures in color images. The main contribution of this segmentation methodology resides in the robust parametric approximations proposed for the multichannel linear prediction error distribution. These are composed of a unimodal approximation based on the Wishart distribution and a multimodal approximation based on the multivariate Gaussian mixture models. For the spatial regularization of the initial class label estimates, computed through the proposed parametric priors, we compare the conventional Potts model to a Potts model fusioned with a region size energy term. We provide performances of the method when using Iterated Conditional Modes algorithm and simulated annealing. Experimental results for the segmentation of synthetic color textures as well as high resolution QuickBird and IKONOS satellite images validate the application of this approach for highly textured images. Advantages of using these priors instead of classical Gaussian approximation and improved label field model are shown by these results. They also verify that the L^*a^*b^* color space exhibits better performance among the used color spaces, indicating its significance for the characterization of color textures through this approach.

Grouping/Degrouping Point Process, a Point Process Driven by Geometrical and Topological Properties of a Partition in Regions

Abstract In the context of image segmentation, we introduce a new kind of point process, called grouping/degrouping point process (GDPP) that aims to aggregate regions from an initial partition of the image according to geometrical and topological criteria. The initial partition, produced by a low-level region-based segmentation process, is represented using a topological map that represents all the geometrical information and topological features of the image partition. Points in the process are localized in regions and newly defined energies of the partition allow to take into account geometrical and topological features like the number of holes or the area of contact between regions. The simulation of the point population is driven by birth and death moves used in a Reversible Jump Markov Chain Monte Carlo method. We propose special birth and death moves using the adjacency relation between regions. Experiments are provided on a sample partition that show the effects of the different potentials. In a 3D medical image, a GDPP based application is provided to segment brain tumor. The results are compared to a region merging approach and to a reference segmentation proposed by an expert. This approach emphasizes the ability of the GDPP to solve real world segmentation problem.

Law recognition via histogram-based estimation

In this paper, we study the problem of recognizing an unknown probability density function from one of its sample which is of interest in signal and image processing or telecommunication applications. By opposition with the classical Kolmogorov-Smirnov method based on empirical cumulative functions, we consider histogram estimators of the density itself built from our data. Those histograms are generated via model selection, more specifically via a codelength-based Information Criterion. From the histograms, we may compute a Kullback-Leibler distance to any theoretical law which is used to complete the recognition. We apply this histogram-based method for law recognition in a theoretical setup where the true density is known as well as in a real setup where data come from radio channel propagation experimentation.

Impedance cardiography filtering using scale fourier linear combiner based on RLS algorithm

The Cardiac Output (CO) can be calculated from the thoracic cardio-impedance signal from several methods, and all of them are linked to the frequency information, information that is limited by the type of filtering used before. A methodology is proposed to evaluate the effect of the commonly used methods of filtering, and an improvement of the SFLC LMS-based algorithm by the use of RLS algorithm is also tested. Performances of algorithms are then evaluated considering different types of noise such as white noise or combination of sinusoidal noises to simulate the effect of respiration and body movements.