Improvement Research and Application of Text Recognition Algorithm Based on CRNN

Abstract

This paper is based on CRNN model to recognize the text in the images of football matches scene, and two improvements are proposed. Considering the edge feature of text is strong, this paper adds MFM layers into CRNN model aiming to enhance the contrast. In order to solve the problem of losing details of image static features in the process of getting contextual features, this paper fuses up these two kinds of features. The training and testing experiments carried out on public dataset and manual dataset respectively verify the validity of the improvements, and the recognition accurate rate is higher than original model.