Using AdaBoost and Combinational Binarization to Position Text Region Automatically for Low-Contrast Image
Gongqin Liu, Murong Jiang, Yaqun Huang, Jianyu Hao, Ke Zeng, Ziyang Zhao · 2018
Locating the text from low-contrast images with appearance variations of text, multi-color is a challenging task which often suffers from position deviation, incomplete region and incorrect etc. In order to solve this problem, a new regional automatic positioning method based on AdaBoost classifier and combinational binarization is proposed. The methodologies include four primary steps: image preprocessing, obtaining the candidate text regions by connected component analysis and AdaBoost, removing background by combinational binarization, expanding and merging text regions by AdaBoost and Similarity. 1933 low-contrast text images were collected for testing the effectiveness. Experimental results show that the proposed method can locate uneven illumination, multilingual content, multicolor of the text more accurately and completely.