Most objects with regular regions could be detected as Maximally Stable Extremal Regions (MSER) [20]. In this paper, We formulate object detection as a bi-label (object and non-object regions) segmentation problem, and propose a graph-based object detection method using edge-enhanced MSER. Specifically, we focus on detecting text in natural images, which is a special kind of object. First, edge-enhanced MSERs are detected as basic letter components; non-text MSERs are then efficiently eliminated by minimizing the cost function which combines both region-based and context-relevant information; and finally, mean-shift clustering is used to group text components into regions. The proposed method is naturally context-relevant, scale-insensitive and readily to be applied on detecting other objects. Experimental results on the ICDAR 2011 competition dataset show that the proposed approach outperforms state-of-the-art methods both in recall and precision. © Springer-Verlag Berlin Heidelberg 2012.
CITATION STYLE
Shi, C., Wang, C., Xiao, B., & Zhang, Y. (2012). Graph-based detection of objects with regular regions. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7508 LNAI, pp. 417–426). https://doi.org/10.1007/978-3-642-33503-7_41
Mendeley helps you to discover research relevant for your work.