Spectral clustering for sonar image segmentation using morphological wavelet and gray level transformation

Spectral clustering algorithm has been used successfully in the domain of image processing except for sonar image segmentation. It cannot capture the sonar target accurately because the sonar image often has ambiguous object edge, extremely complex noisy background and critical shadow impact. In this paper, a new spectral clustering segmentation method base on morphological wavelet and gray level transformation was proposed. Firstly, a morphological midpoint wavelet which can do the image denoising effectively and make the object edge more clearly was constructed to enhance the sonar image; secondly, the gray level threshold transformation was used to remove the shadows of sonar image and the threshold was automatically obtained by the average iteration method; finally, constructed a new spectral clustering segmentation system for sonar image. The simulation experimental results demonstrate that the proposed method was more suitable for sonar image segmentation than the standard spectral clustering segmentation method.