Assessing the quality of a hyperspectral image is a difficult task. However, this assessment is required at different levels of the instrument design: evaluation of the signal to noise ratio necessary for a particular application, determining the acceptable level of losses from compression algorithms for example. It has been shown previously that a combination of five quality criteria can provide a good evaluation of the impact of some degradation on applications, such as classification algorithms for example. This paper refines this concept, providing a representation of the degradation which allows predicting the impact on applications.

This paper shows the interest of an interpolation method based on parametric modeling to retrieve missing samples in ECG signals. This problem occurs more and more with the emergence of telemedicine applications. The different links (fixed access network (PSTN), mobile access network (GSM/GPRS and future UMTS) or satellite interfacing (DVB- RCS technology)) involved in e-health applications are liable to induce errors on the transmitted data. These errors/losses can occur anytime and anywhere (according to the channel availability, memory overflows, protocols, etc) during a transmission process. Therefore the recovering of missing samples for biomedical signals is of great interest. The method used in this paper is based on a left-sided and right-sided autoregressive model, i.e., the interpolation algorithm uses the samples before and after the sequence of missing samples. An objective measure is used to assess the method performance. Results show that this interpolation method represents a really suitable technique to ECG signal reconstruction in a possible corrupted transmission.

Zerotree based coders have shown a good ability to be successfully adapted to 3D image coding. This paper focuses on the adaptation of EZW for the compression of hyperspectral images with reduced complexity. The subordinate pass is removed so that the location of significant coefficients does not need to be kept in memory. To compensate the quality loss due to this removal, a signed binary digit representation is used to increase the efficiency of zerotrees. Contextual arithmetic coding with very limited contexts is also used. Finally, we show that this simplified version of 3D-EZW performs almost as well as the original one.

This paper studies a new spectral analysis strategy for detecting, characterizing and classifying spectral structures of an unknown stationary process. The spectral structures we consider are defined as sinusoidal waves, narrow band signals or noise peaks. A sum of an unknown number of these structures is embedded in an unknown colored noise. The proposed methodology provides a way to calculate a spectral identity card, which features each of these spectral structures, similarly to a real I.D. The processing is based on a local Bayesian hypothesis testing, which is defined in frequency and which takes account of the noise spectrum estimator. Thanks to a matching with the corresponding spectral window, each I.D. card permits the classification of the associated spectral structure into one of the following four classes: Pure Frequency, Narrow Band, Alarm and Noise. Each I.D. card is actually the result of the fusion of intermediate cards, obtained from complementary spectral analysis methods.

Within the frame of machinery maintenance, spectral analysis is a helpful tool. Therefore, an automatic spectral analysis tool, capable to identify each component of a measured signal would be of interest. This paper studies a new spectral analysis strategy for detecting, characterizing and classifying all spectral components of an unknown process. Indeed, any vibration signal can be considered as a mixture of components, a component being either a sinusoidal wave, or a narrow band one. We assume that a sum of an unknown number of these components is embedded in an unknown colored noise. The complete methodology we propose provides a way to feature each component in the spectral domain. The first idea is not to choose one specific spectral analysis method but, rather, to concatenate the results of complementary algorithms. For each one, the noise spectrum is estimated by a nonlinear filter and spectral component detection is managed with a local Bayesian hypothesis testing. This test is defined in frequency and takes account of the noise spectrum estimator. Thanks to a matching with the corresponding spectral window, each component detected is classified into one of the following four classes: Pure Frequency, Narrow Band, Alarm and Noise. The second main idea is then to propose a fusion of the classification results, leading to a complete description of each spectral component present in the signal. This spectral classification is particularly interesting within the context of condition monitoring. Examples are given on real vibratory signals and show the performance of the proposed automatic method, which is particularly well adapted to signals having a high number of components.

La quantité de données produite par les capteurs hyperspectraux nécessite un algorithme de compression efficace qui restè a définir. Les propriétés statistiquesparticulì eres devraient permettre d'obtenir des algorithmes de compression efficaces Etant données ses propriétés et sa faible complexité, la transformée en ondelettes est un candidat prometteur pour la décorrélation des images hyperspectrales. Ce papier propose une méthode pour trouver la décomposition en ondelettes optimale pour les images hyperspectrales et in-troduit la possibilité d'une décomposition non isotropique. La décomposition donnant le meilleur compromis débit-distortion est choisie. Cette décomposition donne de bien meilleures per-formances en terme de débit-distortion que la décomposition isotropique classique. L'inconvénient de cette décomposition optimale réside dans sa complexité importante. Une seconde décomposition, fixe cette fois, est définie et montre des perfor-mances quasi optimales tout en gardant une complexité faible.

Hyperspectral sensors have been of a growing interest over the past few decades for Earth observation as well as deep space exploration. However, the amount of data provided by such sensors requires an efficient compression system which is yet to be defined. It is hoped that the particular statistical properties of such images can be used to obtain very efficient compression algorithms. This paper proposes a method to find the most suitable wavelet decomposition for hyperspectral images and introduces the possibility of non isotropic decomposition. The decomposition is made by choosing the decomposition that provides an optimal rate-distortion trade-off. The obtained decomposition exhibits better performances in terms of rate-distortion curves compared to isotropic decomposition for high bitrates as well as for low bitrates.

This paper studies a new spectral analysis strategy for detecting, characterizing and classifying the different “spectral st ructures” of an unknown stationary process. A “spectra l structure” is defined as a sinusoidal wave, a narrow band signal or a noise peak. The spectral analysis strategy is based on the use of several successive and complementary spect ral analyses. Then, the proposed methodology provides a way to calculate a “spectral identity card” of each spectral struct ure, similarly to a real I.D. card. This I.D. card including all information related to this structure results from th e fusion of intermediate cards, which are obtained from different spectral analysis al gorithms. The I.D. card permits the classification of the detected spectral structur e into one of the following four classes: Pure Frequency, Narr ow Band, Alarm and Reject.

Subband decomposition has been shown to achieve very good performances for frequency estimation, particularly when parametric methods are used. This paper introduces a subband multichannel autoregressive spectral estimation method allowing to exploit the knowledge of intercorrelations between subseries in order to improve frequency estimation performances. This method is detailled then applied to a signals composed by a sum of 2 close sinusoids embedded in noise. Simulation results illustrate the interest of the proposed method.

Le découpage en sous-bandes est réputé pour ses très bonnes performances en matière d’estimation fréquentielle, en particulier lorsqu’on utilise des méthodes paramétriques. Cet article présente une méthode d’estimation spectrale basée sur le découpage en sous-bandes et la modélisation auto-regressive multi-dimensionnelle qui permet d’exploiter la connaissance des inter-corrélations entre les signaux de sousbande afin d’améliorer les performances de l’estimation fréquentielle. Le principe de la méthode est présenté puis appliquée à la résolution de 2 fréquences très proches dans le cas de signaux composés de deux fréquences pures très proches noyées dans du bruit. Des simulations effectuées sur des données synthétiques illustrent les performances de ce nouvel estimateur qui ouvre des perspectives intéressantes.