Low SNR Sonar Image Restoration Based on Mixed Probability Statistical Model in Wavelet Domain
Ping Xia, Qiang Ren, Dongxia Shi, Bangjun Lei, Yaobin Zou, Guang-zhu Xv · 2019
Since it is difficult to get texture details for sonar image denoising with strong noise and weak feature information. A sonar image enhancement algorithm in wavelet domain based on Gaussian mixture model and Gaussian mixture model is proposed to preserve the weak feature information of the sonar image. Under strong noise interference, the traditional enhancement method has certain difficulties in measuring the similarity of the details of the sonar image. For this reason, wavelet multi-scale analysis is performed on the sonar image to extract the weak feature information of each resolution. Secondly, a directed probability map model between adjacent scales in the wavelet domain is constructed to realize the similar weak feature information association. According to the parent node state probability and the transition probability matrix, the state values of the corresponding child nodes are determined, and the relationship between the state information of parent and child node is constructed. Thirdly, the Gaussian mixture model is used to fit the wavelet coefficient state distribution, and the neighborhood coefficient correlation is used to describe the relationship between the weak feature information in the scale. Finally, the wavelet coefficient estimation of the restored image signal is calculated by the state probability obtained in the expectation maximization (EM) algorithm, and the sonar image is reconstructed. The results of the contrast experiment are verified by visual effects and objective evaluation. The proposed algorithm can preserve the weak edge and contour information while suppressing the noise of the sonar image. It has better peak signal-to-noise ratio, signal-to-noise ratio and Structural similarity.