The questions of matching the descriptions at recognizing visual objects in computer vision systems are discussed. The technology of structural recognition which based on the method of tf-idf which used in mathematical linguistics and information retrieval is offered. As a result of perform the conversion of descriptions from variety of the characteristic features of image to the vector of linguistic characteristics that significantly reduces the computational cost. The results of experiments confirming the efficiency and effectiveness of the method are presented.
computer vision, image database, structural image recognition, features, weighting scheme tf-idf, vector description, similarity measure, relevancy