Underwater image segmentation method based on MCA and fuzzy clustering with variational level set

Noise and non-uniform gray level in underwater image cause the low segment precision, this paper presents a segmentation framework based on morphological component analysis (MCA) and fuzzy clustering with variational level set. The framework uses MCA method to sparse decompose the image into texture and cartoon parts. The cartoon part contains the mainly information with less noise, which is good for segmentation. In order to obtain the cartoon part's weaken edges, we have proposed the variational level set method combine with fuzzy clustering. The segmentation algorithm utilizes the subordinate properties of the fuzzy clustering membership function to construct the new variational level set, and uses the minimization of energy functional to get the evolution equation of the level set function. That process can get the final segmentation. The experiment results illustrate that the proposed algorithm, compared with classical segmentation methods, as Fuzzy C-means clustering model, Chan-Vese model and Local Binary Fitting model, can reduce the influence of the noise and obtain a better segmentation result for underwater images.