Sonar Image Denoising via Adaptive Overcomplete Dictionary Based on K-SVD Algorithm

In order to remove the noise of sonar image more effectively, the adaptive overcomplete dictionary based on K-SVD algorithm is carried out in this paper. Given a set of training signals from noisy image, the predefined dictionary is trained so that the new dictionary leads to the best sparse representation for sonar image, but not for the noise. Experiments are provided to demonstrate the performance of the proposed method, as compared with other denoising methods. Results show that this method, which has the capability of adaptation, is particularly appealing in terms of both denoising effect and keeping details, and has improved performance over traditional methods.