Philippe Courmontagne

Speckle noise reduction in SAS imagery

Synthetic aperture sonar (SAS) is actively used in sea bed imagery. Indeed high resolution images provided by SAS are of great interest, especially for the detection, localization and eventually classification of objects lying on sea bed. SAS images are highly corrupted by a granular multiplicative noise, called speckle noise which reduces spatial and radiometric resolutions. The purpose of this article is to present a new adaptive processing that allows image filtering, for both the additive and multiplicative noise case. This new process is based on the marriage between a multi-resolution transformation and a filtering method. The filtering technique used here is based on the two-dimensional stochastic matched filtering method, which maximizes the signal-to-noise ratio after processing and minimizes mean square error between the signals approximation and the original one. Results obtained on real SAS data are presented and compared with those obtained using classical processing.

The Adaptive Stochastic Matched Filter for SAS Images denoising

SAS images are highly corrupted by a granular multiplicative noise, called the speckle which reduces spatial and radiometric resolutions. The purpose of this article is to present a new adaptive processing that allows speckle noise reduction in SAS images, while keeping the spatial resolution. This processing is based on the two-dimensional stochastic matched filtering method with a signal autocorrelation function adaptation and a sub-image size adaptation. Comparison with some classical approaches on real SAS data reveal the efficiency of such an idea

SAS Image De-noising by the Curvelet Stochastic Matched Filter Transform

Over the past few years a great deal of research has been dedicated to SAS image de-noising. Indeed, such images are highly corrupted by a granular multiplicative noise, called the speckle noise, due to the construction of the image itself. This noise reduces spatial and radiometric resolutions. The goal of the research in the area of SAS image de-noising is double: on one hand to reduce the noise level and on the other hand to preserve the spatial resolution. In this context, recently a new approach has been developed: the curvelet transform. Such a method allows good spatial resolution preservation but the results are still slightly disturbed by the introduction of disturbing straight lines due to the process itself. The purpose of this article is to improve the de-noising power of the curvelet transform, while keeping its spatial resolution preservation. To achieve such an objective, the idea is to couple this transform with a filtering technique, known as the stochastic matched filtering method. Results obtained on real SAS data are presented and compared with those obtained using the classical curvelet transform.

Stochastic matched filtering method applied to SAS imagery

SAS (synthetic aperture sonar) has been used in sea bed imagery. Indeed high resolution images provided by SAS are of great interest, especially for the detection localization and eventually classification of objects lying on sea bed. SAS images are highly corrupted by a granular multiplicative noise, called the speckle which reduces spatial and radiometric resolutions. Most of techniques used consist in the use of multilook processing in range, image-domain filters, adaptive filtering or wavelet-domain filtering. The purpose of this article is to present a new adaptive processing that allows image filtering, for both the additive and multiplicative noise case. This processing is based on the two-dimensional stochastic matched filtering method, which maximize the signal to noise ratio after processing and minimize mean square error between the reconstructed signal and the original one. Experimental results on SAS images are presented and compared to those obtained using some classical approaches.

Multiresolution analysis of SAS images

Detection and classification of underwater mines with Synthetic Aperture Sonar (SAS) images is a challenge that can be performed in studying either the echoes or the shadows of mines. However, SAS images present a strong speckle level due to the construction of the image itself. To reduce this speckle level, filtering methods are generally used but all of them strongly deteriorate either the shadow or the echo of the mine. In this article, we propose a new speckle reduction method which allows enhancing jointly mines echoes and shadows. This new process is based on the marriage between a multiresolution transformation and a filtering method. Results obtained on real SAS data are presented and compared with those obtained using classical processing.

An improvement to the pulse compression scheme

The pulse-compression is a technique mainly used in sonar, radar and echography to augment the range resolution as well as the signal to noise ratio. This is achieved using a matched-filtering of the received signal with the bandpass transmitted signal. Taking into account the main assumptions of the matched filter theory, the use of the bandpass transmitted pulse as matched filters impulse response is only available if the useful signal is well known and if the noise is white, which is not the case in practice. For this reason, we propose to improve the classical pulse-compression technique using the stochastic matched filter, which ensures a maximization of the signal to noise ratio, when the useful signal is a realization of a random process and the disturbing signal a colored noise. Results obtained on synthetic and real data are proposed and discussed.

On the use of time-frequency domains for the improvement of the Stochastic Matched Filter Pulse Compression scheme with a high speed computing architecture

In several domains of engineering technologies, such as telecommunication, sonar imaging, positioning systems, radar, medical imaging ..., the main problem is to identify a transmitted useful pulse in a noise-corrupted received signal. A solution to this problem consists in using a modulated pulse for emission and a matched filter for reception. Such a concept is known as pulse-compression. Taking into account the main assumptions of the matched filter theory, the use of the bandpass transmitted pulse as matched filters impulse response is only available if the useful signal is well known and if the noise is white, which is not the case in practice. For this reason, it has been recently proposed an alternative to the classical pulse-compression scheme taking into account the random nature of the useful signal and the coloration of the noise. This new principle is known as the SMF-PC. Although, this method allows a great improvement in terms of signal to noise ratio compared to the classical pulse-compression technique, the SMF-PC appears under optimal due to stationary assumptions. In this context, the purpose of this paper is to propose an improvement of the SMF-PC by coupling this technique with a time-frequency method. Results obtained on synthetic and real data are proposed and discussed.

An improvement of the discriminating capability of the active sonar by optimization of a criterion based on the Cramer-Rao lower bound

This study is a contribution within the framework of detection of underwater targets using active sonar. This paper aims at proposing an original method allowing a great improvement of the sonar performances. This technique uses a filter, determined by optimizing a criterion based on the Cramer-Rao lower bound. Experimental results are proposed and compared with other approaches.

On time-frequency representations for underwater acoustic signal

In order to develop a new human perception inspired process useful for underwater acoustic signal processing, the purpose of this paper is to present a new human physiology based time-frequency representation. This method is based on the use of the Mel filters, classicaly used for the voice recognition. Several experimentations on real underwater signals are presented and discussed.

An improvement on SAS image formation

Synthetic Aperture Imaging Systems are mature and offer a vast choice of reconstruction algorithms. Most of them use pulse-compression. This step is usually performed as a matched-filtering with the bandpass transmitted signal. Taking into account the main assumptions of the matched filter theory, the use of the bandpass transmitted pulse as matched filters impulse response is only available if the useful signal is perfectly known (i.e. deterministic) and if the noise is white, which is not the case in practice. For this reason, we propose to substitute the classical matched filter used for the pulse-compression by its extension to random signal. Results obtained on synthetic and real data are proposed and discussed