基于最小类内方差的相关滤波器研究  

Research on Correlation Filter Based on Minimum Within-Class Variance

作　　者：蒋琦 肖遥 JIANG Qi;XIAO Yao(School of Computer and Software Engineering,Xihua University,Chengdu 610039)

机构地区：[1]西华大学计算机与软件工程学院,成都610039

出　　处：《现代计算机》2020年第13期51-55,共5页Modern Computer

摘　　要：最大间隔相关滤波器(MMCF),作为一种结合支持向量机(SVM)的相关滤波器方法,在检测和识别应用中体现出优异的性能。然而,与SVM相同,MMCF仅考虑边界上的样本点,忽略样本的分布信息。针对这一问题,提出最小类内方差相关滤波器(MCVCF)。与MMCF相似,MCVCF体现大间隔原理。但是,不同于传统方法,MCVCF进一步利用样本的整体分布情况,在训练过程中引入类内的离散度信息,增强其抗噪能力,进而获得更符合真实样本情况的滤波器,提高滤波器的检测分类能力。实验结果表明,MCVCF是有效的,其在目标检测准确率和平均识别率上均优于传统相关滤波器。Maximum Margin Correlation Filter(MMCF),as a correlation filter method that combines the Support Vector Machine(SVM),demonstrates excellent performance in detection and identification.However,same as SVM,MMCF only considers the samples on the boundary,and ig⁃nores the overall distribution information of the samples.Aiming at the problem above,Minimum Class Variance Correlation Filter(MCVCF)was proposed.Similar to MMCF,MCVCF embodies the principle of maximum margin.Different from the traditional methods,MCVCF further utilizes the overall distribution information of the samples.It is added to the scatter information within the class during the training process to enhance its anti-noise ability,thereby obtaining a filter that is more in line with the real sample condition and improving the detection and classification ability of the filter.The experimental results show that MCVCF is effective and superior to traditional corre⁃lation filters in terms of target detection accuracy and average recognition rate.

关 键 词：相关滤波器 支持向量机 类内散度 滤波响应 样本分布信息 

分 类 号：TN713[电子电信—电路与系统]