基于脉冲耦合神经网络的多区域图像分割

为解决传统脉冲耦合神经网络(pulse coupled neural network,PCNN)仅限于二值分割且无法对灰度缓慢变化的大范围区域进行完整分割的问题,提出了一种基于PCNN的多区域图像分割算法。将分割图像经过平滑和归一化后送入PCNN,在快速连接机制作用下,每次迭代处理中具有相似状态的神经元可实现同步点火,完成单个图像区域的完整分割。经过预定的迭代次数后,以各神经元的点火次数为新输入图像各像素点的灰度值,然后经平滑和过归一化后再次送入PCNN重复上述处理,完成多区域图像分割。Berkeley图库的实验结果显示,该算法高效、鲁棒,可有效应用于图像分割。

Multi-region image segmentation based on pulse coupled neural network

In order to solve the problems that the traditional pulse coupled neural network (PCNN) refers only to binary segmentation and does not work well for bigger image regions with sluggish gray variation,a multi-region image segmentation method was proposed based on PCNN. First, the initial image was preprocessed by smoothing and normalizing and put into PCNN. Then, with the help of fast linking, neurons with similar states fired synchronously to finish single region segmentation in each iteration processing. After pre-configured iterations, the total firing times of each neuron were calculated as the pixel intensity of new input image,and then preprocessed by smoothing and normalizing again,and finally put into PCNN. The above processing was repeated to complete multi-region segmentation. Experimental results on Berkeley image database showed that the proposed method was efficient, robust and could be used to segment image effectively.