中国光学, 2016, 9 (5): 523, 网络出版: 2016-10-19
有限离散剪切波域的红外可见光图像融合
Fusion of infrared and visible images based on finite discrete shearlet domain
图像融合 有限离散剪切波 对比度 区域平均梯度 平移不变性 image fusion finite discrete shearlet contrast regional average gradient shift-invariant
摘要
针对目前图像融合过程中的不足之处,结合有限离散剪切波具有高的方向敏感性和抛物尺度化特性,提出了一种有限离散剪切波变换下的图像融合算法。首先对严格配准的多传感器图像进行有限离散剪切波变换,得到低频子带系数和不同尺度不同方向的高频子带系数; 然后对低频子带系数采用全局特征值和像素点之间的差异性与区域空间频率匹配度相结合的融合算法,高频方向子带系数采用方向权重对比度与相对区域平均梯度和相对区域方差相结合的方案; 最后通过有限离散剪切波逆变换得到融合图像。实验结果表明,与其他的融合算法相比较,本文算法不但有良好的主观视觉效果,而且3幅图像的客观评价指标分别平均提高了09%、38%、31%,26%、38%、29%和15%、125%、59%,充分说明了本文融合算法的优越性。
Abstract
Aiming at the deficiency of the current image fusion process, combining with good directional sensitivity and parabolic scaling properties of Finite Discrete Shearlet Transform(FDST), a new image fusion algorithm based on FDST is proposed. Firstly, the registration multi sensing images are decomposed by FDST, and the low frequency sub-band coefficients and high frequency sub-band coefficients of different scales and directions are obtained. The fusion principle of low frequency sub-band coefficients is based on the method of combining the differences between global attribute and each pixel with region spatial frequency matching degree. As for high frequency sub-band coefficients, sum of the directional weight contrast can be adopted as the fusion rule, which combines with the relative region average gradient and relative region variance. Finally, the low frequency information and high frequency information are reconstructed to image by Finite Discrete Shearlet Inverse Transform. The results indicate that the algorithm proposed in this paper has a good subjective visual effect, and its quality indexes has been increased averagely by 09%、38%、31%, 26%、38%、29% and 15%、125%、59% respectively compared with other fusion algorithms, which shows that the algorithm is superior to other fusion algorithms.
陈清江, 张彦博, 柴昱洲, 魏冰蔗. 有限离散剪切波域的红外可见光图像融合[J]. 中国光学, 2016, 9(5): 523. CHEN Qing-jiang, ZHANG Yan-bo, CHAI Yu-zhou, WEI Bing-zhe. Fusion of infrared and visible images based on finite discrete shearlet domain[J]. Chinese Optics, 2016, 9(5): 523.