基于点集不变性匹配的目标检测与识别方法

许多图像目标都可以用其上有代表性的一些点（如角点或边缘上的点）及其相互关系作为特征进行描述，通过模板点集和目标点集之间的匹配可以达到识别的目的，本文提出了一种新的平面点集间的与各自的平移、旋转、尺度变化和排列次序无关的距离描述，并且这种描述在一定范围内具有唯一性，另外在此基础上，给出了一种新的目标检测与识别的算法，该算法不仅能够从较为复杂的背景中检测出经过多个平移、旋转及尺度变化的目标，而且对部分

THE METHOD OF OBJECTS' DETECTION AND RECOGNITION BASED ON MATCHING OF SETS OF IMAGE POINTS WITH INVARIABLE PROPERTY

Many of objects in images can be presented by some important points on them(e.g. corner points or edge points).The recognition task can be completed through matching these points. This paper proposes a new definition of the distance between two sets of image points, this definition has invariant property for rotation, translation, variance of scaling and the order of alignment of these points, and the distance is unique in some ranges.Based on this distance, this paper introduces a new algorithm that can not only detect and recognize objects which may be translated, rotated and scaled variation on complex background,but also shows a good property when part of object being occlusions. So a new effective method is given for the recognition that based on the set of image points.