To address the problem that the traditional multi-focus images fusion methods cannot fully use of spatial context information. A novel image segmentation method for multi-focus image fusion is proposed. This method consists of two main steps, using PSPnet combined, and the region optimization using ConvCRFs for multi-focus image fusion. PSPnet is utilized to extract the focused region of the source image, and ConvCRFs is used to optimize the segmentation map to obtain a refined map. Finally, 20 couples of color multi-focus images are employed as experimental datasets, and the contrast results show that the proposed method has a better fusion visual effect than the other state-of-the-art in subjective and objective point of view.

In underwater scenes, the quality of the video and image acquired by the underwater imaging system suffers from severe degradation, influencing target dete