Machine Vision Inspection Systems, Image Processing, Concepts, Methodologies, and Applications

4Object Descriptor for Machine Vision

Aparna S. Murthy¹* and Salah Rabba²

¹ EIT, PEO, Canada

² Ryerson University, Toronto, Canada

Abstract

In this chapter, we discuss various object identifying techniques in a two-dimensional plane. Object descriptors of different kinds and data structures used to store them are discussed. The comparison of object is based on the constructed data structure. The properties of the object, such as invariance and completeness, are also encoded as a set of descriptors. We discuss the chain code and polygonal approximation, the boundary descriptors for matching the objects. The boundary descriptors are also called contours, whereas moments, Fourier descriptor, and Quadtree are property descriptors used in template matching. The broad classification is that boundary describes the shape of the object, whereas the region descriptors describe the content of the object. Depending on the application, we have to use either boundary or region descriptors for template matching or feature extraction.

Keywords: Polygonal approximation, moments, Zernike polynomial, quadtree

4.1 Outline

Object selection is a tradeoff between performance and accuracy. Particularly, in machine vision, time versus precision for object selection plays a crucial role. With digital images, there are regions of image which are of interest. These regions are a group of segmented pixels that are used for processing. Such regions are often represented by numbers called “object descriptors.” ...