A novel scene text extraction method based on a multi-resolution markov random field

作者：Zeng Zhigao^*; Yi Shengqiu; Wen Zhiqiang; Zhu Yanhui; Zhu Wenqiu

来源：Journal of Computational Information Systems, 2014, 10(21): 9405-9412.

DOI：10.12733/jcis12728

• Markov Random Fields
• Maximum a posteriori estimation
• Monte Carlo Markov chain
• Multiresolution wavelet decompositions
• Scene text extractions
• Spatial informations
• Text extraction
• Wavelet domain

Traditional text extraction methods do not utilize the text context or the spatial information of the Scene text image effectively, so texts are difficult to be extracted validly when the scene background is complex. Therefore, a new method is proposed to solve this problem in this paper. Firstly, the scene image will be processed using multi-resolution wavelet decomposition. Secondly, models of the labeling field and the feature field will be built respectively, and the labeling field model will be trained using the Monte Carlo Markov chain method to estimate model parameters. At last, an inference process will be implemented according to the trained model, that is to obtain the optimal label distribution which can be used to realize the text extraction effiectively, using the maximum a posteriori estimation algorithm.

• 出版日期2014