In this paper is presented one solution to efficient, robust and cheap recognition of artistic pictures on the walls of museums and exhibit halls that reveals satisfactory measure of universality in order to be applied in the areas of trade, process industry, quality control, etc. This solution can be used in a wide range of applications where there is a demand of classifying objects on basis of their visual properties in a large number of existing classes. Here is proposed a method of selective grouping of pattern vectors as training sets for classifiers (artificial neural networks in this case), providing a smaller number of hidden layers in networks, achieving more precise performances and significantly expanding a number of classes to be classified. Selection approach is used in the very classification as well - neural networks are fed with input pattern vectors chosen from subsets determined by additional coefficients. .
展开▼
