Abstract: Segmentation algorithm based on neural networks has high computational complexity and computation load. To resolve this problem, a multi-focus image fusion algorithm was proposed based on the different definition between the focus region and non-focus region of single focal length image. It effectively combined the multiscale, multidirection, anisotropy and shift-invariant qualities of NonSubsampled Contourlet Transform (NSCT) in image decomposition, as well as segmented and fused the low- frequency with clustering of high-frequency. The results show that this algorithm is an effective method in multi-focus image fusion.

Key words: NonSubsampled Contourlet Transform (NSCT), multi-focus image fusion, definition, shift-invariant, region segmentation

摘要： 针对基于神经网络分割算法计算复杂、运算量大等问题，提出一种根据单焦距图像聚焦区域和失焦区域局部相对清晰度的不同进行区域分割的多聚焦图像融合算法。该算法有效结合了非下采样Contourlet变换(NSCT)在图像分解中的多尺度、方向性、各向异性和平移不变性等特点，利用各方向高频分量的聚类来对低频分量进行分割、融合。实验表明该算法是一种有效的多聚焦图像融合方法。

关键词: 非下采样Contourlet变换, 多聚焦图像融合, 清晰度, 平移不变性, 区域分割