A Kernel-Learning-Based Fusion Scheme for Multi-Modal Medical Image Fusion in Shift-Invariant Shearlet Transform Domain

By combing the shift-invariant shearlet transform (SSIT) and the kernel-learning based fusion rule, a fusion algorithm to improve the performance of the traditional multi-scale decomposition (MSD) based fusion methods is proposed. The SSIT is firstly employed to provide better sparse representations for the features; then, the kernel principle components analysis and a support vector machine with the generic multiple kernel learning scheme is constructed to produce the composite SIST coefficients; The final fused results are obtained via the inversion of the SIST. According to the visual comparison on the experimental results, the pseudo-Gibbs phenomenon can be effectively suppressed. Furthermore, the best value of five selected quantitative metrics, whose higher value indicates better fusion results, can be obtained by the proposed method. All the experimental facts demonstrate that the proposed fusion method outperforms the traditional MSD-based fusion methods, such as the wavelet-based, contourlet-based, both visually and quantitatively.