| With the rapid development of big data and artificial intelligence technology,the field of computer vision has received more and more attention;as an important branch in the field of computer vision,the direction of text recognition is even more so.The development of high-performance computing platforms and the emergence of large-scale data sets have made deep breakthroughs in text recognition algorithms based on deep learning in this field.However,many algorithm models have different application scenarios,and the scenarios of each algorithm model are universal weaker.In text recognition tasks,the degree of matching between algorithm model features and image text features plays a decisive role in the accuracy and robustness of the recognition effect.At present,few studies have been carried out on the applicability of multiple text recognition technologies in different application scenarios,thereby limiting the application of text recognition technologies in comprehensive applications such as intelligent transportation systems and autonomous driving.This paper aims to explore the applicability of various text recognition technology solutions to the scenario,and studies each text recognition algorithm.The main work and contributions are as follows:1.In-depth study of text recognition algorithm based on deep learning,analysis of various algorithm models under current mainstream text recognition technology solutions,covering network structure comparison,recognition process and other aspects.2.Summarizes the image text features in the main application scenarios of text recognition,and designs two application scenarios covering different types of text features as research data sets.Through multiple sets of comparative experiments,the scene applicability of each algorithm model is studied and analysis.3.Based on the experimental results of scene applicability of the above algorithm models,the input images were classified by adding a scene classification network,and the corresponding text recognition algorithm was selected for the input in different scenes,thus a universal text recognition system was realized.Through testing,the accuracy of the designed system on both experimental data sets is improved by more than 5%.4.This paper decouples the text recognition system from the perspective of engineering,realizes the fast access of text recognition task under the new scene,and improves the scalability of the system. |
