Object Selection by Grouping of Straight Edge Segments in Digital Images
Object Selection by Grouping of Straight Edge Segments in Digital Images
1
V. Volkov1, R. Germer2, A. Oneshko1, D. Oralov1 Department of Radioengineering, The Bonch-Bruevich State University of Telecommunications, Saint-Petersburg, Russia 2 Institute of Technical Physics e.V., TU Berlin, Germany
Abstract – A new method for finding geometric structures in digital images is proposed. An adaptive algorithm of straight line segments extraction is developed for manmade objects description in digital images. It uses an adjustment of oriented filter angle for precise extraction of line corresponding to real edge. Perceptual grouping approach is applied to these segments to obtain simple and complex structures of lines on the base of their crossings. Initial image is presented as a collection of closed structures with their locations and orientations. Applications to real aerial, satellite and radar images show a good ability to separate and select specific objects like buildings and other line-segment-rich structures. Keywords: object recognition, feature-based image matching, perceptual grouping, content-based image retrieval, building and road extraction
matching, which require the development and investigation of specific object models and feature descriptions with the use of straight line segments [11-24]. Though plenty of works were devoted to theoretical aspects of grouping problems there are not so much practically effective algorithms for manmade object selection in real images [3,15,17,19,20,23-27]. In addition it is often difficult to obtain performance characteristics for such algorithms, choose criteria and make comparative analysis.
2
Related work
1
Introduction
Object extraction, selection and classification are most studied problems of image processing and computer vision. They have important applications for segmentation, visual tracking, image matching, image indexing and image retrieval [1-8]. Model-based approaches instead of
1
V. Volkov1, R. Germer2, A. Oneshko1, D. Oralov1 Department of Radioengineering, The Bonch-Bruevich State University of Telecommunications, Saint-Petersburg, Russia 2 Institute of Technical Physics e.V., TU Berlin, Germany
Abstract – A new method for finding geometric structures in digital images is proposed. An adaptive algorithm of straight line segments extraction is developed for manmade objects description in digital images. It uses an adjustment of oriented filter angle for precise extraction of line corresponding to real edge. Perceptual grouping approach is applied to these segments to obtain simple and complex structures of lines on the base of their crossings. Initial image is presented as a collection of closed structures with their locations and orientations. Applications to real aerial, satellite and radar images show a good ability to separate and select specific objects like buildings and other line-segment-rich structures. Keywords: object recognition, feature-based image matching, perceptual grouping, content-based image retrieval, building and road extraction
matching, which require the development and investigation of specific object models and feature descriptions with the use of straight line segments [11-24]. Though plenty of works were devoted to theoretical aspects of grouping problems there are not so much practically effective algorithms for manmade object selection in real images [3,15,17,19,20,23-27]. In addition it is often difficult to obtain performance characteristics for such algorithms, choose criteria and make comparative analysis.
2
Related work
1
Introduction
Object extraction, selection and classification are most studied problems of image processing and computer vision. They have important applications for segmentation, visual tracking, image matching, image indexing and image retrieval [1-8]. Model-based approaches instead of