Abstract
An approach for decomposition of visual images by clustering and breaking them down into geometric figures is considered. Multilevel hierarchical clustering algorithm to form three emphasized levels of clusters such as rectangles, closed regions and integrated areas is proposed. Advantages of such decomposition in three stages are as follows: images covered by rectangles are planned to be formatted and compressed, image fragments could be taken for the preliminary pattern recognition or could easily be corrected, hierarchically constructed fragments are good material to form pattern features for searching procedures. The algorithm complexity, the proposed approach of scanning searching area to reduce it, the rolling up criteria and key parameters for its control are investigated. The results of pattern analysis by structure features for some practical problems are presented in the article.
Highlights
The clustering methods are widespread for visual patterns segmentation, data mining, indexing and searching. [1−9]
Various approaches are used for improving and decomposition of images, among them being assessment of image fragment local characteristics [8], forming regions of special figures [6] or clustering according to criteria of statistical dependencies [7]
The tasks of the decomposition algorithm when being applied to visual patterns are to solve the following problems: 1) to get full description for all parts of image with its all possible characteristics; 2) to ensure an access for every part of pattern independently of its position on three levels hierarchical tree: integrated areas → closed regions → rectangles
Summary
The clustering methods are widespread for visual patterns segmentation, data mining, indexing and searching. [1−9]. Various approaches are used for improving and decomposition of images, among them being assessment of image fragment local characteristics [8], forming regions of special figures [6] or clustering according to criteria of statistical dependencies [7] These works, as well as a number of others, do not consider the variety of control parameters, by which researcher could investigate different dependences of clustering results, e.g., such as brightness threshold and intensity sensitivity for number of clusters to be formed. Some investigations were held to describe different dependences: number of clusters, parameters of closed regions and integrated areas due to criteria for merging: cluster brightness threshold, algorithm sensitivity or shape and dimension constraints. Presented 3-stages clustering algorithm extracts some pattern structural properties (pattern features) which were taken to analyze and classify images
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
