CAS-DBNet: A Novel Framework for Text Detection

Text in natural scenes often faces problems such as text clutter, different sizes, and slant, which make text detection more difficult. In order to improve the accuracy of Text Detection, based on the Real-time Scene Text Detection with Differentiable Binarization (DBNet), An improved text detection method, CAS-DBNet, is proposed. The core of the improvement is to significantly improve the feature extraction and spatial awareness capability of DBNet network by integrating Coordinate Attention (CA) and S2Attention (S2) modules. Specifically, CA captures the contextual relevance of text area effectively by combining channel information and spatial information. S2Attention further enhances global context awareness by modeling second-order statistics. Experimental results show that the accuracy of the improved algorithm on the test set reaches 92.69%, which is 5.66% higher than that of the original algorithm. The proposed algorithm significantly improves the accuracy and robustness in text detection tasks, demonstrating its superior performance.