基于最小Tsallis交叉熵改进型PCNN图像分割算法

An algorithm of image segmentation based on improved PCNN via minimum Tsallis cross entropy

针对传统脉冲耦合神经网络(PCNN)模型中存在的待定参数过多且难以选择、循环迭代次数难以确定的缺陷,提出了一种基于最小Tsallis交叉熵改进型PCNN图像分割算法.根据改进的内部活动项和阈值衰减函数初始化模型参数,根据图像特性自适应设置链接强度系数和链接权值矩阵;利用最小Tsallis交叉熵准则确定PCNN循环迭代次数,采用双边带滤波器对分割图像进行滤波,获取最优图像分割结果.仿真实验结果表明,该算法提高了分割后图像的区域一致性、区域对比度、形状测度,缩短了运行时间,改进了图像分割效果.

With regard to too many undetermined parameters which are hard to select and difficulty of determining the iterative times in the traditional pulse coupled neural network（PCNN）model,an algorithm of image segmentation based on improved PCNN via minimum Tsallis cross entropy is proposed.Firstly,the model parameters are initialized according to the improved internal activity term and the threshold attenuation function,meanwhile,both the link strength coefficient and the link weight matrix are set adaptively according to image characteristics.Then,the loop iteration times of PCNN are determined using the minimum Tsallis cross entropy criterion.Finally,the segmented image is filtered via bilateral filter to get optimal result.The experiment results show that the uniformity measure,the contrast measure and the shape measure are improved,and the running time is decreased,so the image segmentation effect is improved via the proposed algorithm.