基于非下采样Contourlet变换的自适应医学图像融合算法

Adaptive Fusion Algorithm for Medical Images Using Non-subsampled Contourlet Transform

提出一种多模态医学图像融合算法.用非下采样contourlet变换(non-subsampled contourlet transform,NSCT)将已配准的源图像进行分解,得到低频子带和多层高频子带,然后在各子带中将NSCT系数进行融合.对于低频子带,根据其特性制定规则融合区域能量、互信息、信息熵;对于高频子带,则依据改进的拉普拉斯能量和融合规则,用遗传算法自动优化其待定参数.将融合后的高、低频子带进行NSCT逆变换即可得到融合图像.对灰度和彩色医学图像进行的实验表明,与其他方法相比,用该算法得到的融合图像包含更丰富的纹理信息,视觉效果较好.

An adaptive image fusion algorithm based on non-subsampled contourlet trans- form （NSCT） is proposed for medical images. Source images are first registered and then decomposed to low and high frequency sub-bands using NSCT. The NSCT coefficients in each sub-band are fused. For coefficients in the low frequency bands, a fusion rule based on regional energy, mutual information and information entropy is used. In high frequency bands, sum of modified Laplacian is used. The final image is obtained from the fused sub-images in the low and high frequency bands using inverse NSCT. Experiments are conducted for gray and color images to compare the propose method with previous algorithms. The results show that fused images using the proposed method contains more texture information, and is visually better.