摘要： 为得到更好的融合效果，将特征级融合与像素级融合相结合，利用Contourlet变换(CT)对源图像进行分解，对于近似图像，利用Canny算子进行边缘检测以得到边缘特征图像，再以边缘特征图像作为交叉视觉皮质模型的输入，根据各神经元的点火次数进行融合；对于细节图像，根据区域能量系数矩阵进行融合。通过多聚焦闹钟图像和CT、MRI图像对该算法进行实验，并以熵、互信息和平均梯度作为融合效果的评价指标。实验结果表明，该算法的性能优于传统融合算法。

关键词: 图像融合, 非采样Contourlet变换, 交叉视觉皮质模型, Canny算子, 特征级融合

Abstract: Feature level fusion and pixel level fusion are combined in order to obtain better fusion effect. Contourlet transform is used to decompose the source images into high-frequency subimage and low-frequency subimage. For the low-frequency subimage, Canny operator is used to take an edge detection and get an edge feature image. The edge feature image is taken as the input of intersecting cortical model, the fusion is made according to the number of neurons fire. For the high-frequency subimage, the energy coefficient matrix is used to make a fusion, and multi-focus clock images and CT, MRI images are used to experiment to the proposed method, entropy, mutual information and average gradient are taken as the fusion evaluation index. Experimental result shows the algorithm is superior to traditional fusion algorithm.

Key words: image fusion, Nonsampled Contourlet Transform(NCT), Intersecting Cortical Model(ICM), Canny operator, feature level fusion

中图分类号: 

• TP391.4