Scene text detection with superpixels and hierarchical model

Gang. Zhou,Yuehu. Liu,Zhiqiang. Tian
DOI: https://doi.org/10.1109/ICIP.2012.6467031
2012-01-01
ICIP
Abstract:Scene text detection is a challenging task for the text-based information extraction systems. We present a novel scene text detection method for this task. The images are over-segmented into meaningful perceptron superpixels, and candidate connected components (CCs)are extracted by combining local contrast and color consistency. The non-text components are then pruned by a hierarchical model consisting of three stages in cascade. Experimental results show that our approach is better than other state-of-the-art methods.
What problem does this paper attempt to address?