Multi-Focus Image Fusion via Clustering PCA Based Joint Dictionary Learning

This paper presents a novel framework based on the non-subsampled contourlet transform (NSCT) and sparse representation (SR) to fuse the multi-focus images. In the proposed fusion method, each source image is first decomposed with NSCT to obtain one low-pass sub-image and a number of high-pass sub-images. Second, an SR-based scheme is put forward to fuse the low-pass sub-images of multiple source images. In the SR-based scheme, a joint dictionary is constructed by integrating many informative and compact sub-dictionaries, in which each sub-dictionary is learned by extracting a few principal component analysis bases from the jointly clustered patches obtained from the low-pass subimages. Thirdly, we design a multi-scale morphology focus-measure (MSMF) to synthesize the high-pass sub-images. The MSMF is constructed based on the multi-scale morphology structuring elements and the morphology gradient operators, so that it can effectively extract the comprehensive gradient features from the sub-images. The "Max-MSMF'' is then defined as the fusion rule to fuse the high-pass sub-images. Finally, the fused image is reconstructed by performing the inverse NSCT on the merged low-pass and high-pass subimages, respectively. The proposed method is tested on a series of multi-focus images and compared with several well-known fusion methods. Experimental results and analyses indicate that the proposed method is effective and outperforms some existing state-of-the-art methods.