Journal of Computer Applications ›› 2012, Vol. 32 ›› Issue (03): 885-888.DOI: 10.3724/SP.J.1087.2012.00885

• Typical applications • Previous Articles     Next Articles

New feature description based on feature relationships for gait recognition

XIANG Jun, DA Bang-you, LIANG Juan, HOU Jian-hua   

  1. College of Electronic Information Engineering, South-Central University for Nationalities, Wuhan Hubei 430074, China
  • Received:2011-09-01 Revised:2011-10-31 Online:2012-03-01 Published:2012-03-01

新的基于特征关系表述的步态识别算法

项俊,笪邦友,梁娟,侯建华   

  1. 中南民族大学 电子信息工程学院, 武汉 430074
  • 通讯作者: 项俊
  • 作者简介:项俊(1984-),女,江苏镇江人,硕士研究生,主要研究方向:模式识别、智能监控;笪邦友(1975-),男,湖北仙桃人,讲师,博士,主要研究方向:模式识别、人脸识别;梁娟(1984-),女,陕西渭南人,硕士研究生,主要研究方向:图像处理、视频监控;侯建华(1964-),男,湖北武汉人,教授,博士,主要研究方向:模式识别、智能监控。
  • 基金资助:

    武汉市科技供需对接计划项目(201051824575);湖北省自然科学基金资助项目(2010CDB02001)。

Abstract: In order to carry on the gait recognition fast and efficiently, a new feature relationship based feature representation was proposed in this paper, which utilized nonstationarity in the distribution of feature relationships. Firstly, relative direction between two adjacent edge pixels in 8-neighborhood region was labeled as one of the attributes characterizing relationship, and distance from edge pixel to shape centroid point as the other attribute. Joint probability function of the two attributes was estimated by normalized histogram of observed values. Secondly, Principal Component Analysis (PCA) was adopted for feature reduction. Finally, the nearest-neighbor classifier was adopted for classification. The experimental result demonstrates that the proposed method was used to CASIA gait database, and got the best recognition rate of more than 90%. Feature dimension of the attributes joint probability matrix is reduced from 900 to 240 with relatively lower computational cost.

Key words: gait recognition, feature relationship, feature representation, Principal Component Analysis (PCA), nearest-neighbor classifier

摘要: 为了快速有效地进行步态识别,利用特征关系非平稳分布的统计特性,提出了一种新的基于特征关系表述的步态识别算法。首先,将剪影轮廓相邻像素点间8邻域相对方向标号作为特征关系属性一,将轮廓边界点与中心点间的距离作为特征关系属性二,经直方图归一化处理,得到两种关系属性的联合概率;其次,结合主成分分析(PCA)降维的方法,提取特征主向量;最后,采用最近邻分类器进行识别分类。实验证明,该算法在CASIA步态数据库上,最高达到了90%以上的识别率,而且与传统的特征关系表述步态识别算法相比,关系属性联合概率矩阵维数由900维下降到240维,大大降低了算法的计算代价。

关键词: 步态识别, 特征关系, 特征表述, 主成分分析, 最近邻分类器

CLC Number: