Abstract: In order to achieve image fusion with high quality, a multi-focus image fusion scheme with co-sparse analysis is proposed. First, a co-sparse analysis operator learning model under regularization is developed, for which the analysis operator is obtained with similarity training data. Then, the images to be fused are represented by the salient features extracted with the analysis operator and their corresponding fusion coefficients are obtained with the choose-max rule. Finally, the fused image is reconstructed by minimizing a total variation norm regularized problem. Experimental results show that the proposed method outperforms the sparse synthesis based fusion method; the gain of the proposed method is up to 0.65 in terms of mutual information and 0.1 in terms of Q_ABF.

Key words: computer application, co-sparse analysis, feature extraction, sparse representation, image fusion, multi-focus