Design of Refined Segmentation Model for Underwater Images

Yi Liu,Haofei Li

DOI: https://doi.org/10.1109/ccisp51026.2020.9273503

2020-01-01

Abstract:Image segmentation is a technique to separate the background and the object of study from the image. For underwater image segmentation, the traditional method cannot meet the requirements of complex image segmentation due to its slow correction speed and large error. The image segmentation method based on level set is an image segmentation algorithm based on geometric contour model. This method is more stable than traditional method, simple operation and accurate result. At present, the level set algorithm has achieved good results in medical image segmentation and other non-underwater image segmentation, but the research of underwater image segmentation is still in its infancy. An improved level set algorithm for image segmentation is proposed. In this paper, the characteristics of underwater images are firstly analyzed, and the core principles of curve evolution and level set method are described in detail. Then, an improved level set algorithm is proposed to achieve accurate segmentation of underwater close-up images. In the experimental part, we test and analyze the effectiveness of the algorithm on the underwater image set containing a variety of organisms, and demonstrate the effectiveness of the algorithm on the representative of the jellyfish image with complex texture. At the same time, the segmentation results of this algorithm and Chan-Vese algorithm are compared experimentally. The actual results show that the level set algorithm can effectively complete the fine segmentation of underwater close-up images and has strong robustness to the interference of complex textures. At present, the algorithm is still very sensitive to underwater light interference. In the future, we will continue to improve the work of this paper and try to reduce the sensitivity of the algorithm to light.