Scene Text Detection in Medical Based on Differentiable Binarization

Scene text detection methods based on segmentation have been extensively employed in the area of scene text detection. However, for tasks involving dense text, multi-directional texts and extreme aspect ratios, existing detection way still struggle to deliver desired results. In this article, we introduce an capable attention mechanism and a better post-processing method. Specifically, the initial image pass through the ResNet backbone network, the resulting feature map is enhanced by an attention module that integrates both global and local features to improve the effectiveness of feature extraction. This enhanced feature map is then fed into the FPN (Feature Pyramid Network) for feature fusion, where multi-scale information is combined to produce the final features used for segmentation. Furthermore, the post-processing module using differentiable binarization, enabling the segmentation network to dynamically adjust the binarization threshold. The effectiveness and feasibility of our method are demonstrated through validation on two datasets.