A Systematic Investigation on Different Scene Text Detection and Recognition Method using Deep Learning Techniques

The method of scene text detection and recognition is useful in a variety of applications ranging from multimedia retrieval, traffic monitoring, and navigation for visually impaired individuals to semantic natural scene understanding. Many researchers have worked hard to create algorithms for detecting and recognizing text in photos and movies. The three major areas of scene text recognition are detection, recognition, and end-to-end text recognition systems. Scene text detection and recognition systems have significant challenges due to scene complexity, text diversity, and real-time constraints. The traditional optical character recognition method focuses on recognizing words from image sources but fails to identify multi-oriented and multi-lingual languages. Proposed deep learning approach combined with neural network based scene text detection and recognition method achieves competitive performance. It is critical in complicated scene and large scale dataset detection and identification, multilingual and multi-oriented detection and recognition, real time detection and recognition, and visual understanding. Integrated approach helps to localize, detect and recognize the text in different orientation and multi-language. Also it will reduce the recognition error by employing the advantage of dictionary search method. Proposed approach will provide reasonable performance for curve text, real time video and image files. On several standard text detection benchmarks, it will obtain competitive results in terms of speed and accuracy.