中文摘要：

结合多尺度变换的图像特征，提出了一种基于纹理特征与相关性结构信息的医学图像融合方法。首先对已配准的源图像进行非下采样Contourlet变换，得到低频、高频子带系数。其次考虑人眼视觉对纹理特征的敏感性，提出局部差分计盒维数来统计图像的纹理信息；分析NSCT高频子带兄弟系数间及其父子系数间的强相关性，分别计算出系数间的结构相似度与邻域拉普拉斯能量和，作为高频子带系数间的广义相关性结构信息。然后对低频提出Sigmoid函数自适应融合，对高频采用广义相关性结构信息取大法。最后进行逆NSCT变换得到融合图像。通过灰度与彩色图像融合实验发现，该算法不仅可以保留源图像的边缘信息，还得到较好的客观评价指标和视觉效果。

英文摘要：

According to the characteristics of multi-scale transform, a medical image fusion algorithm based on textural features and generalized correlation structure information is proposed. Firstly, nonsubsampled contourlet transform is conducted for registered source images to get low-frequency and high-frequency sub-band coefficients. Secondly, considering the sensitivity of human eye vision to the textural feature, local differential box counting dimension is used to collect textural information of images, We analyze the strong correlation between brother coefficients and father-son coefficients respectively in high frequency sub-bands of NSCT. Moreover, structure similarity in those coefficients and the sum of Laplace energy between of neighborhoods are calculated as generalized correlation structure information of high frequency sub-band coefficients. Thirdly, Sigmoid function self-adaption fusion is proposed on low frequency sub-bands, and absolute value rule of generalized correlation structure information is presented on high frequency coefficients. Finally, the fused image is obtained by performing the inverse NSCT. It is found through gray level and color image fusion experiment that, the proposed approach can preserve marginal information of source images effectively and improve the objective evaluation index and visual quality.