Analysis for Text Detection on Scenic Images Using Deep Learning

Text detection in scenic images is the process of precisely localizing instances of text in them. This detection is required for a numerous computer vision applications including text recognition, autonomous vehicles, script narrators, translation apps etc. Text detection and localization refer to the process of identifying and locating text within an image or document. Scene text detection was mostly limited to putting together handcrafted low-level attributes of the texts contained in images until the emergence of deep learning in this area. A combination of methods or exploring domainspecific approaches help in enhancing the efficacy of detecting text in scenic images. In this paper, we have implemented the performance of three text detection models EAST, CRAFT and Text Fuse Net on two benchmark public datasets used for detection and localization. Experiments on two benchmark datasets show that the Text Fuse Net achieves state-of-the-art performance. Specifically, we have achieved an F measure of 94.6% and 92.2% on IC-2013 and IC-2015. respectively, followed by CRAFT [89.8%, 84.8%] and EAST [85.3%, 80.7%1.