关键词: 图像文本, 维吾尔文本区域定位, 通道增强MSER, 卷积神经网络, 区域融合算法

Abstract:

In order to locate Uyghur text regions in images accurately and effectively, an image text region location method based on channel enhancement Maximally Stable Extremal Region（MSER） and Convolutional Neural Network（CNN） is proposed. Channel-enhanced MSER is applied to extract candidate regions, non-text and repeated regions are removed according to heuristic rules of text features and CNN classification results, word-level text regions are obtained through a region fusion algorithm, missing text regions are recalled according to the color similarity and spatial relationship of the regions, and the recalled regions are classified and fused through CNN to locate image text regions. The experimental results show that the proposed method can locate text regions accurately and effectively, and has robustness and applicability.

Key words: image text, Uyghur text regions localization, channel-enhanced MSER, Convolutional Neural Network（CNN）, region fusion algorithm