关键词: 关键词: 越南场景文字检测, 声调符号, 模型联合训练, 分割模型, 重复检测

Abstract: This paper presents a text detection algorithm for Vietnamese scenes, based on a modified instance segmentation method Mask R-CNN, aiming at the lack of training data for Vietnamese scene text detection and the incomplete detection of Vietnamese tone markers in the detection. In order to segment Vietnamese scene text with tone marker accurately, only the feature map in P2 layer is utilized to segment the text area, and the mask matrix size of the text area is adjusted to 14 × 28 from 14 × 14 to adapt the shape of most texts. To overcome the problem that duplicate text detection boxes cannot be eliminated by the non-maximum suppression algorithm, this paper designs a filter module for the text areas is added to the detection module, effectively eliminate duplicate detection boxes. This paper uses a model joint training method to train the network. The training process consists of two parts: The first part is the training of The Feature Pyramid Network and The Region Proposal Network of the model. The model use open Latin text data on an enormous scale training to enhance the generalization ability of the model to detect text in different scenes; The second part is the training of the candidate box coordinate regression module and the segmentation module, this part use pixel-level Vietnamese scene text data training to enable the model to segment the Vietnamese text area including tone markers. Many cross-validation experiments and comparative experiments verify that the proposed method has better Precision and Recall under different Intersection over Union (IOU) thresholds compared with Mask R-CNN.

Key words: Keywords: Vietnamese scene text detection, tone marker, model joint training, segmentation module, duplicate detection