光学学报, 2017, 37 (2): 0215003, 网络出版: 2017-02-13
复杂环境下用于人体目标红外图像分割的改进PCNN方法
Improved PCNN Method for Human Target Infrared Image Segmentation Under Complex Environments
机器视觉 红外图像分割 脉冲耦合神经网络 各向异性 拉普拉斯分量绝对和 噪声抑制 machine vision infrared image segmentation pulse-coupled neural network anisotropy sum of modified Laplacian noise suppression
摘要
为了解决复杂环境下红外人体目标分割应用中当前脉冲耦合神经网络(PCNN)方法常出现的噪声适应性差、目标边缘细节模糊等问题,提出了改进的PCNN方法。根据红外噪声特点,利用加权均值滤波和各向异性高斯滤波设计了模型反馈输入域的权值矩阵;采用改进的拉普拉斯分量绝对和表示PCNN的连接强度参数,从而实现了该参数的自适应设置;用点火区域平均灰度值构建动态阈值的方法,实现了PCNN模型的迭代控制。对IEEE OTCBVS和自拍数据库中的250余幅红外人体图像进行对比实验,结果表明,提出的方法能够有效抑制红外噪声,分割出带有较多边缘细节的人体目标,与其他PCNN分割方法相比,该方法还具有较优的平均概率兰德指数和较低的平均全局一致性误差。
Abstract
To solve the problems of poor noise adaptability and blurred edge details of current pulse-coupled neural network (PCNN) methods in the application of human target infrared image segmentation under complex environments, an improved PCNN model is presented. Based on the characteristics of infrared noise, the weight matrix of the feeding input field is designed by the weighted mean value filtering and the anisotropic Gaussian filtering. The improved sum of modified Laplacian is introduced as the linking strength of the PCNN model to set this parameter adaptively. The dynamic threshold is expressed as the average gray value of the fired area to control PCNN iterative process. The proposed method is performed on more than 250 infrared human images from the IEEE OTCBVS database and the self-built database. The experimental results demonstrate that this method can effectively suppress infrared noise and keep many edge details of human targets. Compared with other PCNN segmentation models, the proposed method also shows good average probabilistic Rand index and low global consistency error.
贺付亮, 郭永彩, 高潮. 复杂环境下用于人体目标红外图像分割的改进PCNN方法[J]. 光学学报, 2017, 37(2): 0215003. He Fuliang, Guo Yongcai, Gao Chao. Improved PCNN Method for Human Target Infrared Image Segmentation Under Complex Environments[J]. Acta Optica Sinica, 2017, 37(2): 0215003.