利用直方图及边缘乘积互信息的PCNN图像分割

摘要

脉冲耦合神经网络(Pulse Coupled Neural Network)在图像分割中有很大的应用.其在实现过程时,传统的阈值选取是按等间隔下降依次试出来的,未考虑到图像的灰度先验分布,这种方法确定的分割阈值难以保证全局最佳,影响最终的分割效果.鉴于此,提出了将直方图和PCNN结合的算法,解决了全局最佳阚值的选取问题.同时提出了新的边缘乘积互信息准则用于判断图像分割的效果,不但能很好地利用图像目标的边缘信息,还可以大大降低计算量.实验表明,该算法可以在提高分割精度的基础上,显著地减少分割运行时问,提高分割效率.%Pulse Coupled Neural Network(PCNN) is widely used in image segmentation. The original way of determining threshold is to try out by decreasing equal interval, without considering the prior gray distribution of images. However, this method can't ensure to find out the global optimal threshold, which would affect the effectiveness of results. In consideration of these facts, histogram is introduced to PCNN model, and solves the problem of selecting global optimal threshold. Furthermore, a new rule for optimal segmentation result is also proposed, named contoured product mutual information. Experimental results show that the proposed technique can achieve higher segmentation accuracy, and improve the running efficiency.