摘要
最大间隔相关滤波器(maximum margin correlation filter,MMCF)结合了支持向量机(support vector machine,SVM)和相关滤波器,提高了相关滤波器的鉴别能力。但是,SVM未利用样本结构信息,训练不充分的缺点也被MMCF继承下来。针对这一问题,提出基于数据分布信息的相关滤波器算法,引入样本类内散度及类内局部保留散度进行训练,并进行目标检测和识别实验,从实验结果来看,所提算法是有效的。
Maximum Margin Correlation Filter(MMCF)combines Support Vector Machine(SVM)and correlation filters to im⁃prove the discrimination ability of correlation filters.However,the problem of SVM not using sample structure information is also inher⁃ited by MMCF.In response to this problem,two correlation filter algorithms based on data distribution information are proposed.The sample within-class divergence and locality preserving within-class divergence are introduced for training,then target detection and recognition experiments are performed.From the experimental results,the proposed algorithm is Effective.
作者
蒋琦
周刚
Jiang Qi;Zhou Gang(School of Computer Science,Jinjiang College,Sichuan University,Meishan 620860)
出处
《现代计算机》
2021年第31期25-32,共8页
Modern Computer
关键词
相关滤波器
支持向量机
类内散度
局部保留散度
correlation filter
support vector machine
within-class divergence
locality preserving within-class divergence