A multi-angle shape descriptor with the distance ratio to vertical bounding rectangles

Abstract

In this paper, we proposed a simple and fast matching shape descriptor, a multi-angle shape descriptor with the distance ratio from the contour point to the vertical bounding rectangle(mADR). Shape descriptors are widely used to describe shape features, especially in shape retrieval and shape classification. Contour point is an important feature of shape. The method used in this paper is to find the ratio of distance from contour points to shape vertical bounding rectangle to the average distance. Because vertical bounding rectangle can reflect the position and size information of shape, and can also be used as a feature of shape. In this method, the shape is rotated 360° according to a certain angle respectively, and the vertical bounding rectangle of the shape after each rotation is obtained, then the distance ratio between the contour of all the rotated shapes and the corresponding vertical bounding rectangle is obtained as the shape features. Compared with the traditional shape descriptor based on contour points such as the shape of the distance based on the centroid and contour descriptor (CCD) and based on the shape of the shape context descriptor (SC) has direction characteristic, can distinguish the outline or rough shape similar to the similar shape, have stronger robustness, simpler process and faster matching speed, more importantly, the experiment results show that the shape descriptor has higher retrieval accuracy.