Comparative Analysis on Text Detection for Scenic Images using EAST and CTPN
Prachi Chhabra, Aparna Shrivastava, Zatin Gupta · 2023
The stated goal of this study is to conduct a thorough analysis of a text detection model built with deep learning. Text detection and localisation methods can pinpoint the precise location of a picture's text region. The primary goal of text detection is to locate instances of text within images of beautiful landscapes. Text reading in scene images is a rapidly growing field of study due to its numerous practical applications in areas such as image/video interpretation, visual search, automated driving, and blind auxiliary. The primary goal of this study is to compare and contrast two text detectors, the East text detector and the CTPN Text detector, based on various criteria to determine their relative merits and shortcomings. The goal here is to improve the efficiency of detecting landscapes in photographs. Although deep learning is present in both approaches, their capabilities are different. This comparison aims to find a better method for larger datasets to develop a more sophisticated, accurate, and time-efficient detection method. Both techniques have been reviewed and implemented, illuminating the potential for enhanced and precise detection with the EAST detector. The accuracy and loss were measured against a bounding box drawn around the text in the test image, and both models performed satisfactorily. Comparison. The publicly available ICDAR 2015 dataset is used in this study. It features test images (a total of 500) and training images (1,000).