基于PCNN的图像多区域迭代分割

对传统经典的PCNN网络进行改进，提出一种新的基于PCNN的多区域图像分割方法。去掉原模型中的一些次要参数，突出灰度对分割的影响；根据图像中存在不同灰度变化的特性，分2阶段完成对图像的分割:初次分割和二次分割。初次分割是利用灰度直方图谷底灰度作为动态阈值进行，使动态阈值对分割边界的影响达到最小；二次分割则对初次分割的结果进行细分割，点火区域和非点火区域灰度差较小，其动态链接系数通过循环迭代搜索确定。二次分割迭代进行，从而实现了对整幅图像的完整分割。其实验结果表明，该方法的错误率小于常规的聚类分割算法和GBS算法。  

Research on Multi-region Iterative Segmentation of Image Based on PCNN

A new image multi-region segmentation method based on PCNN is proposed in the paper. The method improves the traditional PCNN image segmentation in such several aspects as following : 1) Removes some secondary parameters in the traditional PCNN model to strengthen the effect on segmentation of image intensity; 2) Divides the segmentation into the initial segmentation and the second segmentation according to the image intensity variation; 3) The initial segmentation is obtained with the bottom value of intensity histogram as dynamic threshold, which make the dynamic threshold impose the minimum impact on the segmentation boundaries; It is subdivided to get the second segmentation. The difference between the firing area and the non-firing area is small and the dynamic linking coefficient is exactly determined by cyclic iteration. The second segmentation is iterated and the complete segmentation of the whole image is performed. The experimental results show that the proposed method is superior to the conventional clustering segmentation algorithm and GBS algorithm.