基于多尺度自卷积归一化直方图的仿射不变量模式识别  

Affine Invariant Pattern Recognition Based on Multi-Scale Autoconvolution Normalized Histograms

作　　者：黄波[1] 赵继印[1] 郑蕊蕊[2] 李敏[2] 

机构地区：[1]吉林大学通信工程学院,吉林长春130025 [2]大连民族学院机电信息工程学院,辽宁大连116600

出　　处：《电子学报》2011年第1期64-69,共6页Acta Electronica Sinica

基　　金：吉林省科技发展计划(No.20090511);中央高校基本科研业务费专项资金(No.DC10010103)

摘　　要：仿射不变量特征提取方法已成为计算机视觉研究的重点课题之一.本文提出一种归一化直方图算法,该算法基于多尺度自卷积变换中密度函数的概念,研究归一化密度函数的方法,构建了从目标图像中提取直方图仿射不变量特征提取算法,实现了基于多尺度自卷积归一化直方图的仿射不变量模式识别.仿真实验表明,本算法对一定范围内的噪声,局部遮挡,照度及视角变化具有良好的适应性,特别是在多种环境下,其识别率优于多尺度自卷积和基于多尺度自卷积的其它直方图算法.Affine invariant feature extraction methods have been one of the key issues of computer vision research.This paper proposes a normalized histogram algorithm based on the density function of multi-scale autoconvolution transform and researches the normalization method of the density function,and constructs a method to extract the histogram affine invariant feature,and implements affine invariant pattern recognition based on multi-scale autoconvolution normalized histograms.Simulation results show that this algorithm has good adaptability for a certain range of noise,partial occlusion,illumination and view angle variance.The recognition rate of this algorithm is superior to multi-scale autoconvolution and other multi-scale autoconvolution based histogram algorithms,especially under a variety of environments.

关 键 词：目标识别 仿射不变特征 归一化直方图 多尺度自卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]