Text Detection Technology for Complex Natural Scenes Based on Improved DBNet

Abstract

Text in natural scenes is often affected by complex text background, irregular text arrangement and other factors, resulting in a significant increase in the difficulty of text detection. In order to further improve the accuracy of text detection, an improved text detection method AC-DBNet for complex natural scenes is proposed based on the segment-based detection technology DBNet. The main points of improvement are to strengthen the feature extraction capability of DBNet network model, add the following extraction module CEM module before the feature map fusion, increase the receptive file through hole convolution, and add the attention guidance module AM to enhance the detection of small target text area, and improve the accuracy of the detection network through these improvements. The accuracy of the improved algorithm in the test set reached 92.1%, which was 0.7% higher than that before the improvement. The improved algorithm improves the accuracy of detection under the task of text detection in real scenes, showing the superiority of the detection algorithm.