NSCT域内结合相位一致性激励PCNN的多聚焦图像融合  

Multi-focus image fusion based on phase congruency motivate pulse coupled neural network-based in NSCT domain

作　　者：刘栋 周冬明[1] 聂仁灿[1] 侯瑞超 LIU Dong;ZHOU Dongming;NIE Rencan;HOU Ruichao(Information College,Yunnan University,Kunming Yunnan 650500,China)

机构地区：[1]云南大学信息学院,昆明650500

出　　处：《计算机应用》2018年第10期3006-3012,共7页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61365001;61463052)~~

摘　　要：针对传统脉冲耦合神经网络(PCNN)无法准确提取多聚焦图像聚焦区域的问题,提出一种利用相位一致性(PC)来检测图像清晰区域,并结合PCNN的多聚焦图像融合算法。首先,利用非下采样轮廓波变换(NSCT)对源图像进行多尺度分解,分别得到图像的高频子带和低频子带;其次,通过计算高频系数的空间频率值(SF)与低频系数的相位一致性值来提取图像高低频子带中的聚焦区域;然后,将SF与PC作为PCNN外部激励来刺激PCNN神经元点火,分别对图像高低频系数进行融合;最后,利用逆NSCT得到最终融合图像。实验采取多聚焦图像Clock、Pepsi和Lab作为三组实验数据集,与传统融合算法及新近提出的几种算法进行对比,所提算法的客观评价参数:互信息、边缘信息度、信息熵、标准差和平均梯度的数值均大于或十分接近于对比算法的最大值;同时从实验结果图与源图像的差值图中可以发现所提算法的差值图包含源图像清晰区域的痕迹明显更少。实验结果表明所提算法能更加准确地提取出图像的清晰区域,更好地保留图像的边缘与纹理等细节信息,得到更好的融合效果。Since the traditional Pulse Coupled Neural Network-based(PCNN)image fusion methods cannot extract the focus region clearly,a multi-focus image fusion technique using Phase Congruency(PC)and Spatial Frequency(SF)combined with PCNN model in Non-Subsampled Contourlet Transform(NSCT)domain was proposed.Firstly,the source images were decomposed into high frequency subband and low frequency subband by NSCT.Secondly,the values of SF and PC were calculated to motivate PCNN neurons to fire to find the focus regions,and then the high and low frequency subbands were fused respectively.Lastly,the fused image was reconstructed through inverse NSCT.Multi-focus image datasets Clock,Pepsi and Lab were utilized as the experimental image sets.In comparison,four classical fusion methods and three newly put forward fusion algorithms were compared with the proposed algorithm.Objective indicators including mutual information,edge intensity,entropy,standard deviation and average gradient were calculated,and the values of the proposed method were greater than or very close to the maximum value of the comparison algorithms;meanwhile,it was clearly found from the difference maps between the experimental result image and the source image that the difference graph of the proposed method contained significantly fewer traces of the clear region of the source image.The experimental results indicate that the proposed method can better extract the clear region of the fused image,and it can better retain details such as edges and textures of the source images,thus,a superior fusion effect is acquired.

关 键 词：相位一致性 空间频率 脉冲耦合神经网络 非下采样轮廓波变换 多聚焦图像融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]