Wavelet-based SAR image enhancement and speckle reduction for classification

Denoising and image enhancement pre-processing techniques are fundamental for segmentation and classification purposes in a wide range of applications. Particularly, in the field of remote sensing, where synthetic aperture radar, (SAR) images are characterized by the intrinsic multiplicative noise, so-called speckle which affects negatively image analysis techniques, such as automatic target recognition, surveillance or environmental monitoring and decreases the global efficiency of classification algorithms. In this paper, we address this problem using wavelet-based image enhancement techniques based on the multiscale edge representation formalized by Mallat and Zhong [1]. In particular, we implement a mother wavelet based on cubic-spline function and apply from modulus-based and phase-based denoising algorithms to linear and non-linear edge stretching contrast-enhancement to improve the image quality and identify the optimum wavelet parameters by analyzing the signal to noise ratio (SNR). Finally, we assess the efficiency of these methods using the Fuzzy C-mean (FCM) clustering algorithm for region classification.