Training deep network models for accurate recognition of texts in scene images

• Main Author: Zhang, Weilun
• Other Authors: Lu Shijian
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/156999
• Institution: Nanyang Technological University

With the advent of the industrial 4.0 era, computer vision and its applications have played an increasingly significant role in the digitization of industries. Images that contain text can be used as sources of information for automating processes. Text recognition from scene images constitutes still a very challenging task because of various backgrounds, variations in size and space, as well as irregular arrangements. The first half of this paper will study and analyse a deep learning model for word recognition in scene images using deep learning. As part of this work, CRNN, one of the most extensively used state-of-the-art deep learning models for text recognition, will be implemented and trained. A number of hyperparameters will be tweaked to obtain optimum performance. In addition, further improvements to CRNN recognition accuracy will be explored. Different image super-resolution techniques and the ImageFilter models will be utilized and compared as a pre-processing stage for recognition in the second part of this paper.