Multi-focus image fusion method based on multi-scale deep supervision

Multi-focus image technology can be used to fuse a fully focused image by using multiple photos taken of the same scene at different focuses. The existing multi-focus image fusion methods have some common problems, such as large computation effort, incorrect estimate of focusing area, unclear fusion boundaries and edge contour loss, etc. To address these problems, an end-to-end fusion framework for multi-focus image fusion is proposed in the paper. Then, to obtain more information from source images, three parallel dilation convolutions are used to extract features from different scales in the paper. In addition, the manually designed fusion rules are replaced by an adaptive attention mechanism in the feature fusion stage. Finally, the final fusion results are reconstructed by convolution modules. The proposed method is better than other existing ones in terms of subjective visual effects and objective quantitative evaluation, according to experimental results.