复杂环境下用于人体目标红外图像分割的改进PCNN方法

发布时间：2018-06-01 00:42

本文选题：机器视觉 + 红外图像分割　；参考：《光学学报》2017年02期

【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]:In order to solve the problems such as poor noise adaptability and fuzzy edge details in the current impulse coupled neural network (PCN-PCNN) method for infrared human body target segmentation in complex environments, an improved PCNN method is proposed. According to the characteristics of infrared noise, the weighted mean filter and anisotropic Gao Si filter are used to design the weight matrix of the input domain of the model feedback, and the improved Laplace component absolute sum is used to represent the connection strength parameters of the PCNN. The adaptive setting of this parameter is realized, and the iterative control of PCNN model is realized by constructing dynamic threshold with the mean gray value of igniting region. More than 250 infrared human images in IEEE OTCBVS and selfie database are compared with each other. The results show that the proposed method can effectively suppress infrared noise and segment human objects with more edge details than other PCNN segmentation methods. The method also has better average probability Rand exponent and lower average global consistency error.
【作者单位】：重庆大学光电技术及系统教育部重点实验室;西南大学电子信息工程学院;
【基金】：教育部博士点基金(20130191110021) 中央高校基本业务费专项资金(XDJK2013C157)
【分类号】：TP391.41

【相似文献】