An efficient sidescan sonar image denoising method based on a new roughness entropy fractal dimension

Abstract

This paper proposed a fractal-wavelet (FW) denoising alternative based on applying texture analysis technique to the fractal matching process. Texture has been regarded as a similarity grouping in an image. Roughness is a perceived property to describe the structural texture. This paper applies the roughness entropy fractal dimension (REFD) algorithm to FW coding process, as the REFD FW algorithm, in finding each range subtree for the optimal matched domain subtree according to the best possible minimal differential of texture similarity measurements. It is believed that such measurement would well capture the texture similarity. The REFD FW algorithm denoises a side-scan sonar image in such a way that the parts of noise-free image have to be approximated as well as possible whereas the noisy parts are discarded. The best possible minimal distance between the two REFD values of domain-range subtrees is used to determine which the best approximation is. The minimal similarity distance quantifies the degree of texture similarity between domain-range subtrees. The REFD FW algorithm have been applied to two side-scan sonar images, one is the wreck of M.V. Sea Angel which is taken by the Polaris, Taiwan, and the wreck of a sailing schooner from MSTL, in different configurations to investigate the corresponding quality of the images using two error criteria: mean square error (MSE) and the peak signal to noise ratio (PSNR). The experimental results indicate that the REFD is appropriate as the criteria of determining range-domain matching in FW coder to well approximate the images. We conclude that the REFD FW algorithm is adaptable in denoising side-scan sonar image and that the images are more appealing visually.