Bharatanatyam hand gesture recognition using normalized chain codes and oriented distances
R. Amrutha, Vandana M. Ladwani · 2016
We propose a system to identify Bharatanatyam hand gestures or Mudras. This involves a preprocessing stage which does a skin based segmentation to obtain the hand boundary. The feature extraction stage involves obtaining the chaincode of the entire contour of the hand followed by normalization. It also includes extracting the Euclidean distance from the centroid to the outermost boundary of the hand along 360 degrees. Extracted features from the training images are used to build four recognition models Naive Bayes, KNN, Logistic Regression, Multiclass SVM. The system shows an accuracy of 88.47%, 87.06%, 89.83%, 92.3% using Naïve Bayes, KNN, Logistic Regression, Multiclass SVM respectively. Multiclass SVM classifier shows the best performance. This system provides an interface for newbies to learn Bharatanatyam Dance form.