卷积稀疏表示和邻域特征结合的多聚焦图像融合

Multi-focus image fusion based on convolution sparse representation and neighborhood features

稀疏表示的分块处理破环了图像的连续性,导致多聚焦融合图像的清晰测度信息严重丢失。针对上述问题,提出了卷积稀疏表示和邻域特征结合的多聚焦图像融合算法。该算法将非下采样轮廓波变换(NSCT)域低频子图通过高斯滤波分解成基础层和细节层,然后选用交替方向乘子算法(ADMM)求解稀疏系数,完成细节层特征响应系数的融合。同时,根据聚焦程度测量函数设计了合理的邻域特征,完成了NSCT域高频子图的融合。实验结果表明:该算法边缘信息传递因子(QAB/F)指标略低于对比算法,但空间频率(SF)、平均梯度(AG)、清晰度(SP)以及视觉信息保真度(VIFF)指标相比于对比算法分别提高了约16.31%、41.87%、19.2%以及12.07%,有效地提取了源图像更深层次的清晰测度信息,克服了稀疏表示的块效应缺陷。

In the image fusion process, the sparse representation has block processing to break the continuity of the image, resulting in serious loss of clear measurement information of the multi-focus fused image. To solve this problem,a multi-focus image fusion algorithm based on convolution sparse representation and neighborhood features is proposed. This algorithm decomposes the low-frequency subgraphs of the non-down sampling contourlet transform (NSCT) domain into the base layer and the detail layer by Gaussian filtering, and then uses the alternating direction multiplier algorithm (ADMM) to solve the sparse coefficients to complete the fusion of feature response coefficients of the detail layer. At the same time, a reasonable neighborhood feature is designed based on the focus degree measurement function to complete the fusion of high-frequency subgraphs in the NSCT domain. The experimental results show that the edge information transfer factor (Q^AB/F) of this algorithm is only slightly lower than that of the comparison algorithm,but the spatial frequency (SF),average gradient (AG),sharpness (SP) and visual information fidelity (VIFF) increase by about 16.31 %, 41.87%, 19.2%, and 12.07 %, respectively, compared with the comparison algorithm. The proposed algorithm effectively extracts the deeper clear measurement information of the source image,also overcomes the blockiness defect of sparse representation, and has better fusion performance.