58. Text Localization in Scene Images Using Faster R-CNN with Double Region Proposal Networks

The problem of text extraction is an interesting area of research in computer vision domain. In the recent years, emergence of various applications on smart hand-held devices such as translation of text from one language to another in real time, computerized aid for visually impaired, user navigation & track monitoring and driving assistance systems, has stimulated the renewed research interest in this domain. Although various Convolutional Neural Network (CNN) based methods have been explored for text localization in scene images, method using Faster R-CNN with double region proposal network (RPN) has not been explored yet. The conventional Faster R-CNN produces regions-of-interest (ROIs) through a single RPN utilizing the feature matrix of the last convolutional layer, whereas the present investigation proposes an end-to-end method of scene text localization where ROIs are generated by double RPNs using the feature matrices of thirteen different convolutional layers and four poolings. Both of these RPNs have then been merged, which enables the system to locate the text regions in the scene images. The performance of the present system has been assessed using ICDAR 2013/2015 RRC test dataset and it has outperformed all existing studies on scene text detection.