Scene Text Recognition using Deep Learning Techniques
Xin Ying, Raja Kumar Murugesan, Siva Raja Sindiramutty, Wei Wei Goh, Sumathi Balakrishnan, Devender Kumar, Sahil Verma · 2024
In the discipline of computer vision, which aims to automatically identify text regions from photos and transform them into editable text data, text recognition in natural contexts presents a difficult task. Deep learning-based methods for text recognition in natural settings have advanced significantly in recent years. The feature extraction-based CRNN technique, which was tested in this paper, performed better. It is possible to accurately portray text properties in real-world environments by using the local entropy property, which is used to describe the information distribution and texture of images. In the CRNN approach based on local entropy feature extraction, the local entropy features from the picture are first extracted using the convolutional neural network. When these features are fed into the recurrent neural network for sequence modeling, the text output is subsequently generated. The CRNN technique based on feature extraction from local entropy may be able to recognize text in real-world situations because local entropy can correctly characterize the texture and information distribution in images. The SCUT-FORU dataset experiment results demonstrate the efficacy of the local entropy feature extraction based CRNN technique for text recognition in realistic settings. There are several fonts, sizes, colors, lighting, backdrops, and other elements in the SCUT-FORU data collection. The CRNN technique based on local entropy feature extraction provides improved accuracy and can perform the text recognition task more quickly, according to experiments on this dataset. In conclusion, the local feature extraction-based CRNN method is a potent text recognition method that may be applied in a range of real-world application scenarios. The continued improvement of deep learning technology will enable future developments and innovations in the field of character recognition in natural settings.