Multidimensional Co-occurrence Matrices for Object Recognition and Matching

摘要: A novel method is proposed for object recognition and matching. It based on the automatic search of features that characterize a certain class using training set consisting both positive negative examples. Special multidimensional co-occurrence matrices are used description representation some basic image structures. The extracted from elements this matrix express quantitatively relative abundance elementary structures, i.e., they quotients matrix. Only which discriminate classes clearly used. demonstrated in numerous applications, falling under general problems texture recognition, defect detection, shape recognition.