基于PCNN的图像直方图均衡化增强  

Image histogram equalization enhancement based on PCNN

作　　者：张煜东[1] 吴乐南[1] 李铜川[2] 王水花[1] 

机构地区：[1]东南大学信息科学与工程学院,南京210096 [2]航天科工集团35所军事代表室,北京100013

出　　处：《东南大学学报（自然科学版）》2010年第1期64-68,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(60872075);江苏省自然科学基金资助项目(BK2007103);东南大学优秀博士学位论文基金资助项目(YBJJ0908)

摘　　要：为了更好地增强图像,提出一种新的图像增强方法.处理分为2个阶段,首先局部增强阶段,利用PCNN模拟空间掩盖效应去除了人眼无法察觉的双边缘,同时在神经元模型中引入侧抑制来模拟Mach带效应,使边缘处灰度差值更大,平滑区域灰度差值更小.其次全局增强阶段,将灰度信息与空间信息耦合到神经元的内部活动项,将阈值设置为局部增强后的图像直方图的累加密度函数,通过比较内部活动项与累加密度函数,得到最终的增强图像.理论与实验均证明了最终图像满足直方图均衡化的要求,不仅对灰度层损失问题免疫,而且直方图近似均衡.In order to enhance images more effectively, a novel enhancement strategy is presented, which is processed by two stages： local enhancement stage and global enhancement stage. In local enhancement stage PCNN （pulse coupled neural network） is used to simulate spatial concealment effect and abnegate the double-edge which is difficult for human eyes to observe. Meantime lateral inhibition is introduced to simulate Mach band effect, which can enlarge the difference of bilateral gray values of edges and can smooth the fiat zones. In global enhancement stage, both gray value in- formation and spatial information are coupled into the inner activity item, and the threshold of the corresponding neuron is set as the cumulative density function of the histogram of local enhanced im- age. Thus through comparing the inner activity item and the cumulative density function, final en- hanced image can be attained. Both theory and experiments demonstrate that this method can equalize the given image perfectly, and not only it is immune from traditional gray scale loss problem, but also its histogram is better equalized than traditional methods.

关 键 词：图像增强 人类视觉系统 脉冲耦合神经网络 

分 类 号：TN911.73[电子电信—通信与信息系统]