Segmentation of Graphical Objects as Maximally Stable Salient Regions
Su Yang, Yuanyuan Wang · 2011
Symbol segmentation is the key to affect the performance of symbol recognition in natural scenes. As experimentally confirmed, MSER is effective in segmenting text but not applicable to segmentation of graphical objects like traffic signs. We propose a color space graphical object segmentation method. It extracts stable region of interest by applying different color similarity threshold to evaluate the stability of proximity among pixels. Experimental results show that it outperforms MSER in segmenting graphical object.