摘要
形状识别在计算机视觉中具有十分重要的意义,理论上矩不变量特征在图像平移、伸缩、旋转时均保持不变.文中提供了针对光学图像上的飞机、汽车目标识别方法的研究,此方法的特点是采用五个不变矩的指数和两个二阶矩为特征,采取直接根据特征值范围进行分类的方式,对18种类型飞机的235幅图片和2种类型汽车的13幅图片进行了识别,识别准确率达到了94%.
Shape recognition is a very important problem in computer vision. Invariant moments are independent of position, scale and orientation in theory. This paper offers a way of recognition of aircrafts and automobiles in the optic images, which used exponents of 5 moment invariants and two secondorder moments as features, and it directly divides the targets by their ranges of the features. Discern to 2：35 pictures of 18 kinds of aircrafts and 13 pictures of 2 kinds of automobiles, the rate of accuracy reaches 94%.
出处
《武汉理工大学学报（交通科学与工程版）》
2006年第4期696-699,共4页
Journal of Wuhan University of Technology（Transportation Science & Engineering）
基金
国家"十.五"规划项目资助(批准号:41307050802)
关键词
计算机视觉
形状特征
矩不变量
目标识别
computer vision
shape feature
moment invariant
targets recognition
