A Hybrid Model Proposal to Recognize the Hand Gestures of Bharatanatyam mudras

Bharatanatyam is one of the oldest forms of Indian classical cultural dance. Mudras are used to convey information visually through hand gestures. The objective of the proposed article is to use models built using AFIS, SVM, and CNN classification to identify the gestures performed by the dancer poses. The datasets required for this task are created by collecting images of Bharatanatyam dancers from various age groups depicting the eight mudras that includes four single handed mudras (Paataka, Mushti, Kapittha and Kataka muha) and four double handed mudras (Anjali, Swastika, Pushpaputa and Garuda). The images are pre-processed using median filtering and Histogram equalization techniques. This is followed by segmenting the hand region from the images of the dancers by finding the exterior boundaries of the hand and removing all connected components. Edge features but are extracted using the edge detector and the SVM classification is used to design the first model based on the edge characteristics to determine the posture. The second model is created using the Discrete Cosine Transform (DCT) features extracted from the images and the ANFIS classifier is used to classify the features leading to recognize the mudras depicted in the images. The third model takes the segmented images itself as input and fed in deep learning CNN to recognize the inputs. The three models are compared using recognition accuracy as the performance measure and CNN is found to outperform the SVM and ANFIS classifiers in hand gesture recognition in Bharatanatyam mudras. This work can be used in creating a framework where an interested learner can acquire or sharpen skills in Bharatanatyam dance form through self-learning itself.