www.cafetinnova.org Indexed in Scopus Compendex and Geobase Elsevier, Geo-Ref Information Services-USA, List B of Scientific Journals, Poland, Directory of Research Journals ISSN 0974-5904, Volume 09, No. 05 October 2016, P.P.2172-2181 #02090548 Copyright ©2016 CAFET-INNOVA TECHNICAL SOCIETY. All rights reserved. Sub-Graph Matching-Based Building Changes Detection Using High-Resolution Remote Sensing Images FENG-HUA HUANG 1-5 , ZHENG-YUAN MAO 3-5 AND WEN-ZAO SHI 3-5 1 Postdoctoral Programme of Electronic Science and Technology, Fuzhou University, Fuzhou 350116, China 2 Yango College, Fuzhou 350015, China 3 Key Lab of Spatial Data Mining & Information Sharing of Ministry of Education, Fuzhou University, Fuzhou 350002, China 4 National Engineering Research Centre of Geospatial Information Technology, Fuzhou University, Fuzhou 350002, China 5 Spatial Information Engineering Research Centre of Fujian Province, Fuzhou University, Fuzhou 350002, China Email: fenghuait@sina.com Abstract: High-resolution remote sensing (HRRS) images of urban regions have large viewing angle variations, significant noise jamming, and obvious building shadows. Hence, deviation and distortion usually occur to the buildings in HRRS images collected at different phases (multi-temporal HRRS images). The traditional detection methodology is ineffective for accurate and efficient building changes detection in multi-temporal HRRS images. In order to address these problems, this paper proposes a sub-graph matching-based building changes detection (SGMBCD) scheme. First, this paper presents a Graphcut-based buildings extracting (GCBE) method from multi-temporal HRRS images. Next, a sub-graph matching-based registration (SGMR) method is devised to register the previously extracted buildings from multi-temporal HRRS images and to obtain matched ASIFT feature point pairs and singular points. Finally, singular points and overlay analysis-based method (SPOA) is developed to detect building changes in multi-temporal HRRS images. The types of building changes included in this paper are changes (e.g., erection, dismantling, repairing, and reconstruction) and non-changes. In order to demonstrate the effectiveness of the proposed SGMBCD scheme, it is compared with five typical algorithms (i.e., BCDBPM, SDBBCD, BCDBICO, NCUTBCD, and RSMBCD) on three sets of multi-temporal WorldView2 test images. Experimental results show that in comparison with the other methods, SGMBCD can effectively address the challenging problem of building changes detection in multi-temporal HRRS images. The average recall ratio, precision ratio and F value is 91.47%, 86.49% and 88.91% respectively, and the average time consumption is 60.3 s. This demonstrates SGMBCD can detect building changes in multi-temporal HRRS images accurately and efficiently. Keywords: High-Resolution Remote Sensing Images, Sub-graph Matching, Building Changes Detection, Overlay Analysis. 1. Introduction Building changes detection via remote sensing images refers to the technique of detecting building changes on the ground using multi-temporal remote sensing data. Building changes include reconstruction and extension due to land use and coverage variation, as well as collapse and damage due to natural disasters [1]. Building changes detection is of great significance in urban planning, GIS data upgrading, smart cities, and military surveillance [2,3]. Currently, the data source that underlies building changes detection is primarily high-resolution remote sensing (HRRS) images. Although HRRS images provide the benefit of building recognition, the images are subject to large angle variations, significant noise jamming, and apparent building shadows. Deviations and distortions usually occur to the same building in multi-temporal HRRS images, which makes it more difficult to detect building changes in HRRS images than those of other objects on the ground, such as water bodies, vegetation, and roads. Other factors that added to the difficulty of building changes detection include: lack of directly relevant 3D data, possibility of the same object exhibiting different spectrums, diversity in building appearance, and complexity of the scene surrounding buildings in urban areas[1- 4].As a result, traditional methods cannot accurately and efficiently detect building changes in multi- temporal HRRS images. Fruitful studies have been done worldwide on this issue. Typical algorithms include the probabilistic model method [5], shadow analysis method [6], inter-class overlay analysis method [7], graph segmentation method [8], and image matching method [9]. These algorithms have shortcomings in terms of self-adaption, accuracy and efficiency, especially for HRRS mages with complex backgrounds. In order to address these problems, this paper proposes a sub-graph matching-based building