Özet:

Abstract Context. Intensive development and expansion of the application possibilities of modern computer vision systems requires in-depth research and creation of more efficient and versatile visual information processing methods. The main problems are related with the research and improvement of information recognition technology in an integrated feature space with regard to the descriptions in the form of image point set descriptors (SURF-features), as well as the necessity of recognition performance estimation in a practical applications. Objective. Article is focused on the research of possibility of constructing and evaluating the effectiveness of median processing models to perform structural recognition of objects in the image in terms of obtaining a compression data in feature space of image database. Method. Transformation of space structural features into vector space in order to increase the speed of the recognition process was proposed. Median processing of descriptions to form ordered finite list of descriptors was proposed as transform method. The result is the creation of method to form and calculate the relevance of image descriptions in the transformed feature space. Results. Implementation of median characteristics analysis to form vector representation allowed to reduce the amount of computing costs significantly and improve recognition performance. Recognition time in comparison with the traditional approach is reduced hundreds of times preserving required efficiency. Simulation and experimental research of the proposed recognition method on the test dataset was performed on the basis of SURF descriptions. Effectiveness in terms of performance is confirmed, comparative evaluation of the quality of recognition for a variety of treatment options is obtained. Conclusions. Perspective properties recognition systems in the space of the structural features of images are systematized. The median analysis allows to reveal new patterns in initial information to provide effective fast recognition. Scientific novelty of research is the method of structural image recognition by applying the median analysis to form compressed vector representation of set of descriptors in the structural description of an image. Move to the vector-listed view considerably improves performance by simplifying recognition processing. Application value of the work is to provide practical programming models for the modification of structural method for detection and confirmation of the effectiveness of the proposed approach in the specific datasets. 

Anahtar Kelimeler: