Structural Tensor Nonlinear Complex Diffusion-Based Method for Sonar Image Denoising

This paper presents a sonar image denoising method based on structural tensor and nonlinear complex diffusion (STNCD). The complex diffusion in STNCD utilizes the imaginary value generated by the diffusion process as the directional structure indicator, then controls the diffusion coefficient by introducing the eigenvalue of the structural tensor and the spatial variation structure factor related to the eigenvector. As a result, the complex value of STNCD ensures that the phase information of the image is preserved during the noise suppression process. Furthermore, we make use of the difference between the principal eigenvectors of the structural tensor to describe the local coherence of the features, and then solve the problem that the phase angle cannot be accurately estimated when constructing the structural tensor. Experimental results show that STNCD not only suppresses sonar image noise but also enhances the signal of image edge, which has a wide range of applications.