A Multi-focus Image Fusion Method Based on Cascade CNN Networks

Due to the fact that the optical lenses have limited depth of field, it is difficult to capture an image in which all the objects are in focus. Multi-focus image fusion is a popular technique to solve this problem. In this paper, a method based on cascade CNN networks (Pixel Network and Block Network) is proposed. The Block Network develops the rough fusion mask and the Pixel Network refines edges of the mask. Then, the mask is processed by rolling guidance filter. Finally, the fusion image is generated according to the mask. Experiment shows that our method achieves better result than the methods being compared. All the code of this article can be downloaded on https://github.com/zhanghexie/multi-focus-image-fusion.