Multifocus image fusion algorithm based on sparse theory and FFST-GIF
1. Pujiang Institute, Nanjing Tech University, Nanjing, Jiangsu 211200, China; 2. School of Information Engineering, Nanjing Audit University, Nanjing, Jiangsu 211815, China
Abstract： To solve the problems of fuzzy detail features and poor comprehensive effect of multifocus images after sparse theory fusion, an image fusion algorithm was proposed based on sparse theory and fast finite shearlet transform with guided image filtering (FFST-GIF). FFST was adopted to disassemble the highfrequency and lowfrequency subband coefficients from the original image. The relative standard deviation algorithm of guided filtering was used to fuse the highfrequency coefficients with rich detail information. The K singular value decomposition (K-SVD) method was used to train the complete dictionary for fusing the low-frequency coefficients by combining the sparse theory. The fused high and low frequency subband coefficients were refused through inverse FFST to obtain a new fused image. Based on MATLAB, the brain MRI images in Harvard University database were selected as samples, and the four objective evaluation indexes of average gradient (AG), spatial frequency (SF), mutual information (MI) and edge preserving information transfer factor (QAB/F) were used to compare the proposed algorithm with the multi focus image fusion algorithms based on non-subsampled contourlet transform and pulse coupled neural network. The fusion experiments in different transform domains and different fusion algorithms were completed. The results show that the proposed algorithm has outstanding advantages in objective comprehensive evaluation index and visual effect. The parameters of each evaluation index are greatly improved with AG and
QAB/F up to 0.081 3 and 0.793 5, respectively, and the proposed algorithm has good application prospect.
