摘要
建立了一套对图像融合质量进行定量评价的方法和准则。其中,对于相同传感器源的图像,利用标准参考图像,选取均方误差、峰值信噪比、参考图像,以及融合图像的信息熵差、互信息作为客观评价指标;对于不同传感器源图像,选取融合图像的标准差、空间频率、信息熵以及融合图像与两源图像的互信息之和作为客观评价指标。应用此方法和准则对基于离散小波框架、基于可操纵方向金字塔变换和基于双树复小波变换三种平移不变性的图像融合方法进行了客观定量评价。实验结果表明,基于可操纵方向金字塔变换的融合方法效果最好,然后是基于离散小波框架的融合方法;基于双树复小波变换的融合方法其性能与前两种接近,且其冗余性最小。
A set of quantitative evaluation methods and rules on image fusion quality were proposed, For the source images coming from the same sensor, the selected evaluation criteria using standard reference image were the mean square error, peak of signal-to-noise ratio, reference image, information entropy difference and mutual information. While for the source images coming from the different sensors, the selected evaluation criteria were the standard deviation, spatial frequency, information entropy and the mutual information between fusion image and source image These evaluation methods were used respectively to evaluate several shiff-invariant fusion approaches, which were based on discrete wavelet frames（DWF）, steerable pyramid（SteerPYR） and dual tree complex wavelet transform （DT-CWT）. The experimental results show that the SteerPYR fusion method is the best, then is the DWF method. The performance of DT-CWT method is close to that of the former two methods and with tess redundance.
出处
《中国惯性技术学报》
EI
CSCD
2007年第6期702-706,共5页
Journal of Chinese Inertial Technology
基金
河南省教育厅自然科学基础研究计划项目(2007510023)
关键词
图像融合
小波变换
小波框架
可操纵方向金字塔变换
双树复小波变换
image fusion
wavelet transform, wavelet frames
steerable pyramid transform
dual tree complex wavelet transform
