Please use this identifier to cite or link to this item:
Title: Дослідження модифікацій методу встановлення релевантності зображень об’єктів за описами у вигляді множини дескрипторів ключових точок
Authors: Гороховатський, В. О.
Keywords: структурні методи розпізнавання зображень
детектор BRISK
кластеризація у просторі дескрипторів
узагальнений дескриптор
релевантність описів
Issue Date: 2018
Citation: Дослідження модифікацій методу встановлення релевантності зображень об’єктів за описами у вигляді множини дескрипторів ключових точок / В. О. Гороховатський, А. А. Васильченко, К. П. Манько, Р. П. Пономаренко // Системи управління, навігації та зв’язку. –2018. – №5 (51). – C. 74–78.
Abstract: INVESTIGATION OF THE RELEVANCE IMAGE OBJECTS ESTIMATION METHOD MODIFICATIONS WITH DESCRIPTIONS IN THE FORM OF KEYPOINTS FEATURES SET V. O. Gorokhovatskyi, A.A. Vasylchenko, K.P. Manko, R.P. Ponomarenko The subject of the paper is the models for estimation of the relevance degree between images in the space of key points descriptors for the implementation of visual images structural recognition methods in computer vision systems. The goal is the experimental modeling of methods modifications implementations effective in terms of performance for estimation of keypoint descriptors similarity based on the bit data analysis approach. The tasks include the development of mathematical and software data processing models for calculation of the structural descriptions similarity, the investigation of the properties and application features of these models, the effectiveness evaluation according to specific images processing results. The methods are to be used: BRISK detector for forming of key point descriptors, data mining, k-means clustering method, methods of bitwise processing and data entry frequency calculation, the theory of bit data hashing, experimental modeling. Following results are obtained. Image classification methods based on the similarity of key point descriptors are improved and applied using the implementation of the bit data analysis approach. The cluster descriptions representation allows not only to reduce the processing time but also to show the sensitivity of method modification to insignificant image feature and its ability to be widely used in computer vision systems. Hashing the description without losing data is significantly (hundreds of times during modeling) accelerates the process of descriptions relevancy degree calculation. The selected hash function can influence the result and help to increase the level of image distinguishing. The construction of the general description in a form of a common descriptor significantly reduces the computing time, because of which the requirement of a prior description processing in order to form a shortened description from the list of valuable descriptors occurs. Conclusions. The contribution of the paper is to improve the structural image recognition method based on the description as a set of key point features using clustering approach, the identification of generalized properties and data hashing to determine the modified relevance measures of the analyzed and etalon descriptions. The practical significance of the paper is the achievement of a significant increase of image relevance calculation speed, confirmation of the effectiveness of proposed modifications on sample images, obtaining of an application software models for research and implementation of classification methods in computer vision systems. Keywords: structural image recognition methods, BRISK detector, clustering in descriptor space, generalized descriptor, hashing, relevancy of descriptions, voting, Hamming metric, speed of relevancy determination.
Appears in Collections:Кафедра інформатики (ІНФ)

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.