基于非下采样Contourlet变换的 SVM多聚焦图像融合

提出基于非下采样Contourlet变换的支持向量机(SVM)多聚焦图像融合算法. 采用非下采样Contourlet变换分解图像得到不同频域子带系数. 针对直接取系数绝对值最大融合规则不能反映图像区域的缺点,提出SVM分类系数融合规则. 根据各子带系数物理意义将区域方差、区域能量作为SVM核函数参考量来选择清晰像素点系数,根据融合系数重构得到融合图像. 结果证明该算法能有效并准确地融合图像中的信息.

Support Vector Machine Based on Nonsubsampled Contourlet Transform for Fusing Multi-Focus Images

The nonsubsampled Contourlet transform (NSCT) provides a shift-invariant directional multiresolution image representation, which leads to a NSCT with better frequency selectivity and regularity. A new approach is improved to fuse multi-focus images with support vector machine(SVM) based on NSCT. The features from the NSCT coefficients are used and SVMs are trained to determine whether coefficients from the source image with the best focus should be used. The kernels of SVMs are improved by using region variance and region energy. The fused NSCT coefficients are used to reconstruct fused image.Experimental results show that the proposed method fuses multi-focus images effectively and accurately.