Recognize objects with three kinds of information in landmarks

Abstract

Object recognition is one of the most important tasks in computer vision domain and it has been extensively applied in many areas during the last three decades. In this paper, a method is proposed to recognize objects in shape spaces in which landmarks have three kinds of information, i.e., landmarks position, angle and feature vector. By comparing angle and feature vector, a landmark in an image can correspond with a few number of landmarks in another image. Then, the pixel's positions of landmarks from two images are used to calculate the Procrustean distance in a shape space. If the value of the Procrustean distance is smaller than a predefined threshold, the two sets of landmarks will be considered to match each other. Applying the proposed method, two-dimensional objects including occluded objects can be recognized quickly.