Natural image character recognition method and system based on deep neural network

Abstract

The invention discloses a natural image character recognition method and system based on a deep neural network. The method comprises the steps of image acquisition and fusion, character positioning, character recognition and rationality judgment. The system comprises an image acquisition and fusion unit, a character detection unit, a character recognition unit and a rationality judgment unit. Thecharacter detection unit automatically locates the region of the characters in the image through the character detector obtained by training. The character recognition unit firstly classifies the characters into three categories: Chinese, English and numbers through the constructed multi-interface CNN and then recognizes the characters of the corresponding category. The rationality judgment unit fuses the output result of the recognition unit with the prediction result of the character-based LSTM model so as to improve the rationality of Chinese continuous character recognition. According to the natural image character recognition method and system, the characters in the natural image can be intelligently recognized, and the core processing model is composed of the deep neural network andcan simulate the human brain to the maximum extent, and thus the robustness and the recognition accuracy are high.

Description

technical field [0001] The invention belongs to the technical field of target character recognition, and more specifically relates to a natural image character recognition method and system based on a deep neural network. Background technique [0002] With the continuous development of modern science and technology and the widespread popularization of the Internet, digital image information resources have also shown exponential growth, and a large amount of image information needs to be manually recorded in work and life. Therefore, how to quickly and accurately recognize the text information that needs to be recorded in the natural image directly through the computer has become an urgent problem to be solved. [0003] The current Optical Character Recognition (OCR) model is mainly divided into two parts: character segmentation and character recognition. However, traditional character segmentation methods need to classify positive and negative samples through the selection o...

AI Technical Summary

Problems solved by technology

[0003] The current Optical Character Recognition (OCR) model is mainly divided into two parts: character segmentation and character recognition. However, traditional character segmentation methods need to classify positive and negative samples through the selection of sliding windows and a large number of SVM calculations. These methods Usually the amount of calculation involved is very large, and it is impossible to complete the online test in real time

On the other hand, traditional character recognition methods solve the problem of feature extraction by manually selecting text features, and such artificially selected features usually have errors in the accuracy and comprehensiveness of features. Not only that, the pixel information of natural images Missing or lighting problems will also lead to a decrease in recognition accuracy

Images