Processing and Recognition of Characters Image in Complex Environment

Abstract

Aiming at the recognition in complex backgrounds using the Optical Character Recognition (OCR) technology, a model with high detection and recognition accuracy for small texts and codes in the images is proposed in this paper. Before recognition, the Retinex image enhancement and median filtering are performed to weaken the influence of ambient lighting and to enhance the image features. The Otsu threshold segmentation is adopted to segment the small text images from the background. Finally, the Differentiable Binarization (DB) algorithm is applied to detect the characters and the Convolutional Recurrent Neural Network (CRNN) algorithm is employed to recognize the detected characters. The experimental results show that in the life scenes, the workpiece components scenes, and the Chinese characters scenes, the recognition accuracies of the proposed model are 95.6%, 98.4%, and 90.9%, respectively; and the recognition times are 0.78s, 0.59s, and 0.63s, respectively. Overall, the average recognition accuracy of the model proposed in this paper reaches up to 94.9%, and the average recognition time is only 0.67s, which verifies the effectiveness and the advancement of the model.