基于最小交叉熵的改进的简化脉冲耦合神经网络的图像分割算法

Image Segmentation Based on Modified Simplified PCNN Model and Minimum Cross-entropy

将简化的脉冲耦合神经网络模型改进,对神经元的连接强度和该模型中的常数动态门限衰减时间常数改进,实验得出固定输出的放大系数,简化了模型的参数个数,进而该简化模型的运行速度降低。并且将此改进的模型与最小交叉熵准则结合,将其应用于图像分割当中。通过计算机仿真实验表明,基于最小交叉熵的改进的简化脉冲耦合神经网络模型分割图像,对比于自动阈值分割,在图像上更能够保留原有的有效信息,并且去除边缘无效模糊信息。通过三种判定分割图像优劣的方法,相同像素的百分比、区域一致性参数三种判定图像分割优劣的方法对比可以得出,基于最小交叉熵的改进的简化脉冲耦合神经网络模型对图像进行分割所得的数值比自动阈值分割法的数值更加理想,从而可以看出,该方法优于自动阈值分割法。

The simplified pulse-coupled neural network model was improved,and the connection strength of neurons and the constant dynamic threshold decay time constant in the model were improved.The amplification coefficient of the fixed output was obtained by the experiment.The number of parameters of the model was simplified,and the running speed of the simplified model was reduced.The improved model is combined with the minimum cross-entropy criterion to apply it to image segmentation.Computer simulation experiments show that the improved simplified pulse coupled neural network model based on minimum cross-entropy can better retain the original effective information on the image and remove the invalid fuzzy information on the edge compared with the automatic threshold segmentation.Through the comparison of three methods to determine the quality of the segmented image,the percentage of the same pixel and the regional consistency parameter,it can be concluded that the value obtained by the improved simplified pulse coupled neural network model based on minimum cross-entropy is more ideal than that of the automatic threshold segmentation method.Thus,it can be seen that the proposed method outperforms the automatic threshold segmentation method.