基于结构相似性的多源图像融合

Multi-source Image Fusion Based on Structural Similarity

对源图像进行小波分解,得到图像分解后的高频系数和低频系数,并且计算源图像的结构相似性。对于低频系数采用系数绝对值取大法进行融合;对于高频系数依据相似性大小对其进行融合,如果相似性大于设定阈值,则高频系数采用系数绝对值取大法,否则采用计算像素点边缘强度法进行融合,对融合后的系数进行小波逆变换得到融合图像。实验结果表明,该方法有效地综合了源图像中的重要信息,融合图像边界清晰,得到更好视觉效果和更优的评价指标,融合效果优于对比实验中的小波方法。

A novel fusion algorithm of multi-source image is proposed based on structural similarity.Orginal images are decomposed by wavelet transform and wavelet coefficients are obtained at different scales.Meanwhile the data structural similarity is calculated.For the low frequency components,,the largest absolute value method is adopted.The selection of high frequency components was obtained by the SSIM.If the SSIM is greater than the threshold,the largest absolute value was used for the high frequency components,or the high frequency components is decided by edge intensity.Simualtion results demonstrate that the reconstructed images in our method outputforms wavelet.This algorithm which obtains both in visual quality and objective evaluation index is efficient to integrate important information from the original images.