基于显著计算与自适应PCNN的图像融合方法  

Image fusion method based on saliency computation and adaptive PCNN

作　　者：杨秀林[1] 黄硕[2] 邓苗[1] 张基宏[1,3] 

机构地区：[1]深圳大学信息工程学院,广东深圳518060 [2]信阳职业技术学院数学与计算机科学学院,河南信阳464000 [3]可视媒体处理与传输深圳市重点实验室(深圳信息职业技术学院),广东深圳518029

出　　处：《山东大学学报（工学版）》2014年第2期35-42,共8页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(61271420);广东省自然科学基金资助项目(S2012020011034)

摘　　要：针对多尺度变换的图像融合对低频系数进行简单的加权平均处理时,不能很好地保护源图像中的显著信息的问题,提出一种将视觉显著计算的结果作为自适应脉冲耦合神经网络的链接强度,通过脉冲耦合神经网络指导多尺度图像融合中低频系数融合的方法。首先对源图像进行形态非抽样小波分解,得到低频系数和各尺度的高频系数,对低频系数采用显著计算与脉冲耦合神经网络的融合规则,高频系数选取绝对值较大者,最后通过反变换得到融合图像。实验结果表明,该方法在一定程度上保留了源图像中的显著信息,改善了互信息、信息熵、平均梯度和边缘保持度等融合指标。Simple average processing of low-pass subbands was usually adopted in multi-scale transform based image fu-sion,which could not protect the saliency information in source images very well. To solve this problem,an image fu-sion method using the saliency computation to drive adaptive pulse-coupled neural network（PCNN）was proposed to optimize the low-pass subbands fusion. First,source images were decomposed by morphological un-decimated wavelet, and low-frequency coefficients and high-frequency coefficients were obtained. Second,low-frequency coefficients were fused by rule based on saliency computation and PCNN,and high-frequency coefficients were selected by strong abso-lute value. Finally,the fusion image was got by inverse transform. Experimental results indicated that saliency informa-tion of source images was obtained to some extent and the fusion indicators,such as mutual information,entropy,aver-age gradient and edge preservation degree were improved.

关 键 词：图像融合 形态非抽样小波 显著计算 自适应 多尺度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]