阐述了小波变换、遗传算法、模糊数学、神经网络、数学形态学等生物细胞图像分割算法以及边缘检测、区域分割等传统图像分割算法为主的生物细胞图像分割技术的发展现状，指明了生物细胞图像本身具有的复杂性、多样性、各自差异性等属性是实现生物细胞图像全自动分割的难点，只有彻底结合生物视觉特性数学模型算法的研究和应用，才能使生物细胞图像全自动分割成为可能。

90年代发展形成的脉冲耦合神经网络（PCNN）模型特别适合于图像分割，边缘提取等方面的应用研究，但众所周知，PCNN模型图像分割效果不但取决于PCNN模型中各个参数的合理选择，而且同时还取决于循环迭代次数的确定选择准则，通常循环迭代次数N的选择通过人工交互方式来确。定正因如此选择合适的准则来确定N是PCNN图像分割的关键，但目前还没有文献提出一个合适的准则来解决这个问题。本文结合图像统计特性和PCNN参数模型提出了熵值最大准则。该准则实现了PCNN神经网络的自动图像分割。对于PCNN的理论研究和实际应用具有非常重要的现实意义。

植物胚性细胞定量分析研究需要首先将其切片图像分割处理，然后进行大分子量化分析但植物细胞切片图像上表现出来的植物细胞特有的复杂属性，使得一般图像分割分析方法很难奏效20世纪90年代中期发展起来的脉冲耦合神经网络PCNN直接来自于动物视觉特性研究，应当适合这类植物细胞图像的分割处理但因目前理论很难解释PCNN数学模型参数与图像分割效果之间的关系，一般较好图像分割效果的获得需多次实验选择这些模型参数同时在模型参数选定的情况下，其循环迭代次数直接关系到分割结果的好坏，而分割好坏的判定需人眼观察分析，这样便引入了人为干预为此提出一种建立在分割图像熵值最大原则上的PCNN植物细胞图像自动分割新算法。

Pulse Coupled Neural Networks(PCNN) is a new Neural Networks which was developed and formed in the 1990's. The key point of PCNN is modulated coupled mechanism, while coupled results produce internal activity. The output of PCNN is binary image sequence, which can be considered the results of threshold segmentation. In this paper, the matrix made by internal activity is regarded as a breadth of image, then which can be conjoined with the technique of traditional threshold segmentation. The application of minimum cross-entropy criterion in the technique of image segmentation makes the discrepancy of information content between image segmented and image after segmentation to be least. A kind of novel algorithm of image segmentation setting on cycle iterations automatically is put forward, after traditional PCNN threshold segmentation mechanism improved with the combination of minimum cross-entropy criterion. Theory analysis and experimental results all show that the best segmentation output can be drew from the simple and sophisticated image using this new algorithm.

Median filter can inhibit the impulse noise in the image, butit always erodes or dilates the edges of images. H.S.Ranganath mentioned that impulse noise could beremoved through modifying the intensity of those contaminated pixels step by step using PCNN. Obviously this method consumes much more time in computation. Combining the PCNN model with the median filter, this paper presents an impulse noise filter based on a simplified PCNN model which has less parameters. Not only can it remove the impulse noise effectively, but also it keeps the details of images as can as possible. It can be verified through experiments and theory analysis that this kind of filter is superior to the normal median filter and the filter mentioned by H.S.Ranganath, no matter in the aspect of noise removal or in the aspect of keeping details.