Blind Estimation of Single Look Side Scan Sonar Image from the Observation Model

The presence of multiplicative speckle noise in side scan sonar images unfavourably affects the detection and classification of ob- jects in the image. This noise is related to the sensing system and the image formation process. Speckle removal is a pre-processing step required in such images for applications like segmentation and registration. Most of the speckle reduction techniques intro- duces a bias after the homomorphic approach is applied on the observation model of the side scan sonar image to convert the multiplicative model into an additive one. In this paper an unscented Kalman filter based approach for single look side scan sonar image estimation from the observation model of the image is proposed. First the noise level in the single look side scan sonar image is estimated using a patch based algorithm and then non blind denoising is done. For the blind denoising part, the heterogeneity of side scan sonar image patches is exploited. Simulation results are given to substantiate the effectiveness of the despeckling method. The different non reference image performance evaluation measures are used to assess the proposed method.