Efficient Estimation of Visual Object Relevance During Recognition Through Their Vector Descriptions

The paper is devoted to some aspects of structural descriptions matching in course of visual object recognition in computer vision systems. The suggested here structural recognition technique employs both the "tf-idf" method used in the information retrieval and the median processing of descriptions. This technique was successfully applied to transform a description consisting of a set of image features into a vector of features or a list of descriptors, which essentially reduces the computational costs. The experimental results confirmed the adequacy and efficiency of the suggested techniques.