Survey on Automatic Script Identification Techniques
Miral V. Donda, Harshadkumar B. Prajapati, Vipul K. Dabhi · 2019
Script identification is an essential task especially in India because of existence of 13 different scripts for writing 22 languages. Major applications of script identification are document sorting, automatic translation, selecting of OCR (Optical Character Recognition) and text area identification. Traditionally, researchers have used feature based methods for script identification but it can be automated through deep learning techniques to reduce comprehensive time. Moreover, CNN (Convolutional Neural Network) based deep learning approaches are less explored for script identification problem. Intention behind this survey on script identification is to make researchers more perceived about the usefulness of deep learning techniques and more specifically, the significance of CNN to deal with script identification problem. This paper contains a useful survey on both feature-based and deep learning based domains on popular foreign and Indian scripts. We have shown comparative analysis between both domains which clearly concludes that deep learning methods could be more useful as compared to feature based methods. In addition, limitations and usefulness of CNN based best practices are also presented.