Uyghur Text Localization with Fast Component Detection

Text localization in image often is an important part of image content analysis and has broad application prospects. Even though there have been many researches focus on it, fast Uyghur text localization in complex background images is still a challenging task. The obstacles mainly come from the huge extracted candidates and the heavy computation of non-text classification. In this paper, we propose a fast framework for Uyghur text localization which handle above obstacles with two effective measures. One is that we propose a stroke-specific detector based candidate extraction scheme. Compared with the common used I-MSER detector, the presented scheme not only produces 2 times less components but also runs in twice faster. The other is a component similarity based clustering is raised, which neither need the component-level classification nor the extra computations. The experiments confirm that our method has achieved the state-of-the-art on UICBI-500 benchmark dataset and runs in near real-time. The localization results also prove that the proposed method is robust to Chinese and English.