A Deep Learning-Based Approach to Single/Mixed Script-Type Identification

Script identification acts as the precursor in recognizing the scene text by optical character recognition (OCR). But this is not the fundamental issue to the OCR engine. Before script identification, script-type classification is necessary since nowadays the scene texts in natural images are not comprised of a single script but also multi-script words at character level are found very frequently in different places like posters, T-shirts graffiti, hoardings, banners, etc. In this work, a deep learning-based framework has been designed to classify single/mixed script images. To assess the effectiveness of the structure presented, experiments were also performed with an outlier class consisting of a wide variety of single scripts. Experiments were performed with over 4 K images and The best precision of 98.30% was reached. This method was compared with a standard deep learning and handcrafted feature-based technique where the proposed technique produced a better result.