Multi-focus Image Fusion Combining Focus-Region-level Partition and Pulse-Coupled Neural Network
Kangjian He,Dongming Zhou,Xuejie Zhang,Rencan Nie,Xin Jin
DOI: https://doi.org/10.1007/s00500-018-3118-9
IF: 3.732
2018-01-01
Soft Computing
Abstract:Multi-scale transforms (MST)-based methods are popular for multi-focus image fusion recently because of the superior performances, such as the fused image containing more details of edges and textures. However, most of MST-based methods are based on pixel operations, which require a large amount of data processing. Moreover, different fusion strategies cannot completely preserve the clear pixels within the focused area of the source image to obtain the fusion image. To solve these problems, this paper proposes a novel image fusion method based on focus-region-level partition and pulse-coupled neural network (PCNN) in nonsubsampled contourlet transform (NSCT) domain. A clarity evaluation function is constructed to measure which regions in the source image are focused. By removing the focused regions from the source images, the non-focus regions which contain the edge pixels of the focused regions are obtained. Next, the non-focus regions are decomposed into a series of subimages using NSCT, and subimages are fused using different strategies to obtain the fused non-focus regions. Eventually, the fused result is obtained by fusing the focused regions and the fused non-focus regions. Experimental results show that the proposed fusion scheme can retain more clear pixels of two source images and preserve more details of the non-focus regions, which is superior to conventional methods in visual inspection and objective evaluations.