激光与光电子学进展, 2015, 52 (12): 121001, 网络出版: 2015-12-08
改进内部活动项的多通道PCNN 彩色图像分割 下载: 601次
Color Image Segmentation Based on Improved Internal Activity Multi-Channel Pulse Coupled Neural Networks
图像处理 多通道脉冲耦合神经网络 内部活动项 耦合平均 最大信息熵 image processing multi-channel pulse coupled neural networks internal activity coupled averaging maximum entropy
摘要
为了充分利用图像彩色信息,克服传统单通道脉冲耦合神经网络的图像分割过程中信息丢失,提出了一种多通道图像分割方法。采用RGB 颜色空间,为每一个色彩分量建立一个输入通道,形成包含三个输入通道的多通道脉冲耦合神经网络,将内部活动项修改为各输入通道耦合平均,动态阈值变化采用上升指数,各通道的连接加权系数矩阵选取三维欧氏距离倒数矩阵。以最大信息熵作为评价标准,通过标准彩色图像进行实验分析,选取多通道脉冲耦合神经网络最佳参数。实验结果表明,多通道脉冲耦合神经网络的彩色图像分割方法对图像的细节信息保留明显,最大信息熵相对平均提高3%,在提高了图像分割效果的同时降低运行时间80%。
Abstract
In order to make full use of the image color information and overcome the traditional single channel pulse coupled neural network information loss in the process of image segmentation. A multi-channel image segmentation method is proposed, input channel of each color component is established for RGB color space. So multi-channel pulse coupled neural networks are formed which contains three input channels. Internal activity is modified based on coupled averaging of each input channel, dynamic threshold changes with exponential ascent, each component of the three-dimensional euclidean inverse distance matrix is calculated as the connection weighting coefficient matrix for each channel, and maximum entropy is adopted as evaluation criteria. Experiments are carried out based on standard color images, optimal parameters of multi-channel pulse coupled neural networks are selected according to test results. Experimental results show that more particulars of color image are preserved by color image segmentation based on multi-channel pulse coupled neural networks. Average value of maximum entropy increases by 3% relatively,image segmentation effect is improved while cost time is reduced more than 80%.
王蒙军, 郭林, 王霞, 郝宁. 改进内部活动项的多通道PCNN 彩色图像分割[J]. 激光与光电子学进展, 2015, 52(12): 121001. Wang Mengjun, Guo Lin, Wang Xia, Hao Ning. Color Image Segmentation Based on Improved Internal Activity Multi-Channel Pulse Coupled Neural Networks[J]. Laser & Optoelectronics Progress, 2015, 52(12): 121001.