Kathakali character identification – Using deep learning techniques and web technologies for Indian cultural heritage

Abstract

Technology advancements have proven to benefit every part of our life. It has shown visible results in all domains including medicine, retail, governance, entertainment to quite a few. There are several ways technology has been used for art and culture. In this paper, we examine the use of technology for a 400 year old art form Kathakali, one of the most sacred arts of Indian Cultural Heritage. Kathakali is known for hand-gestures, dance elements, music and sophisticated costumes and makeups. Kathakali is salient part of Indian Cultural Heritage. However, there haven't been much work to promote this part of Indian history. Through this paper, we aim towards taking one major step between our technological field and Indian cultural heritage. We use artificial intelligence (AI) techniques such as convolutional neural networks for Kathakali character identification from images. We employ different deep learning techniques for the same and evaluate the performance of our algorithms. We observe that through rigorous parameter tuning and experiments, up to 97.5% accuracy has been obtained. We have personally recorded images of Kathakali from live performances over 20 years. The data set developed for this study contains images for 6 Kathakali characters - Pacha, Kathi, Minukku-Female, Red Beard, White Beard and Kari-Male and have been made available for the public to use in future. We then develop a web platform wherein users can upload new Kathakali images and the platform uses the AI algorithm behind to identify its character type. The web platform has been designed to deploy in a cloud environment so that it can benefit the public. We show how technology advancements can help improve audience reach for such old Indian art forms through this paper.