Text Sequence Recognition in Natural Scenes Based on Deep Recurrent Neural Networks

Abstract

Detection and recognition of variable length text sequence in natural scene has a wide range of applications, such as license plate recognition, road sign recognition, advertising and so on. Traditional way of text information extracting and recording are manually operated, which is high cost, low efficient and inconvenient. To solve this problem, CNN, RNN and CTC networks are combined to be a union detection and recognition system. Experiment results show that this system can accurately and efficiently detect and recognize of variable length text sequence in English in real scene.