Multidimensional Co-occurrence Matrices for Object Recognition and Matching

✍ Scribed by Vassili Kovalev; Maria Petrou

• Publisher: Elsevier Science
• Year: 1996
• Tongue: English
• Weight: 591 KB
• Volume: 58
• Category: Article
• ISSN: 1077-3169
• DOI: 10.1006/gmip.1996.0016

✦ Synopsis

occurrence matrices and gray-level run length matrices, A novel method is proposed for object recognition and match-suggested by Haralick et al. [10] and Galloway [11], have ing. It is based on the automatic search of features that characbeen widely used for texture description, texture classificaterize a certain object class using a training set consisting of tion, and segmentation [12-14]. Haddon and Boyce [15both positive and negative examples. Special multidimensional 17] have renewed the interest in co-occurrence matrices co-occurrence matrices are used for the description and repreby proposing the use of special co-occurrence matrices for sentation of some basic image structures. The features are exedge detection and estimation of the optic flow field, and tracted from the elements of this matrix and express quantitaby using co-occurrence matrices as look-up tables to emtively the relative abundance of some elementary structures, phasize differences between the salient (atypical) and the i.e., they are quotients of certain elements of the matrix. Only background (typical) image features. features which discriminate the classes clearly are used. The method is demonstrated in numerous applications, falling un-We argue that the attributes used to represent an object der the general problems of texture recognition, texture defect are problem dependent and that often what characterizes detection, and shape recognition.