特征匹配度结合边缘检测的图像融合技术

Image Fusion Technology Based on Feature Matching and Edge Detection

针对传统方法中使用单特征不足以衡量图像像素清晰度的局限性,利用非下采样轮廓波变换(non-subsampled Contourlet transform,NSCT)的系数特点和局部特征匹配度,结合基于区域分割的边缘检测算法,提出一种针对多聚焦图像的融合算法。该算法首先通过NSCT变换将两幅待融合的源图像分解为一个低频分量和一系列高频分量;其次,针对低频分量包含了源图像的大部分能量和信息的特点采用局域信息熵。局域改进的拉普拉斯能量和的统计特征进行特征匹配度融合,以及对高频分量中包含了源图像的细节纹理信息的特点采用区域平均梯度的兄弟关联权重进行融合;最后对源图像的高频分量进行边缘检测加权平均融合,将边缘图覆盖到经NSCT逆变换的初步融合图像上,得到最终融合图像。将所提算法与传统NSCT变换方法和DWT变换方法进行对比,该算法在视觉效果和平均梯度、空间频率、标准差与互信息多个评价指标上都有较好的结果。

In view of the limitation that single feature is not enough to measure the image pixel sharpness in traditional methods,using the coefficient characteristics and local feature matching degree of non-subsampled Contourlet transform(NSCT),together with the edge detection algorithm based on region segmentation,we propose a multi-focus image fusion algorithm.Firstly,two source images to be merged are decomposed into a low frequency coefficient and a series of high frequency coefficients by NSCT transformation.Secondly,for the feature that the low-frequency component contains most of the energy and information of the source image,the local information entropy,the local improved Laplace energy and the statistical feature are adopted for feature matching fusion,and for the feature that the high-frequency components contains the detail texture information,the fusion is conducted by the regional average gradient sibling weights.Finally,the edge detection fusion with weighed average of the original image high-frequency components is performed,and the fused edge map is overlaid onto the NSCT inverse transformed preliminary fusion image is carried out to obtain the final fused image.Compared with the traditional NSCT and DWT,the proposed algorithm has ideal effect in visual effect,mean gradient,spatial frequency,standard deviation and mutual information.