Scene text detection based on hierarchical multilayer perceptron

In this paper, a new scene text detection method based on hierarchical multilayer perceptron (MLP) is proposed. First, connected components (CCs) are segmented locally by text probability map. Then, a novelty hierarchical architecture consisting of two MLP classifiers in tandem is utilized to analysis the CCs. In this hierarchical setup, the first stage MLP classifier is trained using unary property features. The second stage MLP classifier is trained for CCs pairs including both posterior probabilities estimated by first stage and relationship features. Finally, candidate text CCs are grouping into words. Experimental results evaluated on the public dataset show that our approach yields better performance compared with state-of-the-art methods.