Classification of expressions in Indian Classical Dance using LBP
Manjeeta R. Kale, Priti P. Rege · 2019
Face analysis has been a topic of interest since the past few decades and a lot of research has been done in the domains of face detection-recognition and facial expression recognition. Amongst the numerous applications that use face analysis, the application that is not much explored is an analysis of expressions in Indian classical dance. The expressions that are peculiarly used in Indian classical dance strictly follow the ancient literature on drama and dance. Also, the expressions are deliberate as opposed to real-life expressions. This is the motivation behind carrying out a separate analysis for dance-specific expressions. The dataset of facial images with dance-specific expressions is created. The images of trained classical dancers with and without makeup are captured in a controlled environment. The local binary pattern, a widely used texture descriptor, is used for feature extraction. The classification of the expressions into 7 classes viz. happy, sad, angry, fearful, surprised, disgusted, and neutral is carried out using linear SVM with 81.38% recognition rate. The expressions in Indian Classical Dance that are difficult to differentiate from a neutral (due to the nature of expression) need to be identified. To do so, the differentiability of the expression from neutral face is calculated using template matching and SVM. This experiment is carried out on our dataset successfully and more than 90% recognition rate is achieved for most of the expressions. "Anger" is identified as the expression that is difficult to differentiate from neutral in the dance specific dataset.