Research on key technology of Chinese text localization in natural scenes

Abstract

To solve the problem of Chinese text localization in natural scenes, a text localization method based on the feature of Chinese characters is proposed. Firstly, the Maximally Stable Extremal Regions (MSER) algorithm is adopted to extract candidate text regions. Then, the separated Chinese character strokes are fused by using mathematical morphological operations. Then, the non-text regions are filtered through heuristic rules and structural features. The Stroke Width Transform algorithm is used on the candidate text regions, and the non-text regions are removed by stroke width features. Finally, a text line construction algorithm in any direction is adopted to merge the localized text regions so as to achieve the final localization of Chinese text. The experimental results showed that the proposed method could provide better results for the localization of Chinese text in natural scenes.