Video character end-to-end detection and recognition method based on deep learning

Abstract

The invention discloses a video character end-to-end detection and recognition method based on deep learning, and belongs to the technical field of video character processing. The method comprises the following steps: performing image size normalization processing on each video frame image of a video sequence segment to be recognized, so that the pre-processed image size is matched with the input of an end-to-end character detection and recognition network; and sequentially inputting the preprocessed images into the end-to-end character detection and recognition network, and obtaining a character recognition result of the video sequence segment to be recognized. According to the invention, end-to-end detection and recognition of video characters are realized, inherent defects such as error accumulation caused by inconsistent multi-module targets are avoided, and the engineering complexity is reduced. In addition, the network structure is optimized through a shared feature extraction network, and meanwhile, a feature map with a relatively large receptive field is input into the recognition branch of the network, so that the feature map can contain information in a larger range compared with input by using an original map, and the recognition accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of video text processing, in particular to an end-to-end detection and recognition method for video text based on deep learning. Background technique [0002] In recent years, with the digitalization of social informatization and the widespread dissemination of multimedia information, how to extract information from massive videos and images has become an urgent problem to be solved. Video text detection and recognition technology can well meet the needs of a large number of video content extraction and review. Compared with manual work, the use of video text detection and recognition technology for video content extraction and review can greatly improve efficiency and reduce labor costs. [0003] Among them, text detection refers to the use of text detection algorithm to detect the input image, determine whether the input image contains text, if the image contains text, then further locate the position whe...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a method for end-to-end detection and recognition of video text based on deep learning, and to improve the existing two-stage video text detection and recognition method through a shared feature extraction network. When the detection result is inaccurate, it is easy to interfere with the recognition result. problems, while improving the efficiency of network reasoning

Images