Wavelet Based Entropy Features for Facial Expression Recognition

Texture features are explored in an active manner for precise classification of facial expressions. In this paper, we inquire into wavelet entropy features obtained from the daubechies (db4), coiflets (coif4) and symlets (sym4) wavelet filters and propose a new novel method to classify seven human facial expressions by employing Viola-Jones face detection algorithm, gamma intensity correction (GIC), discrete wavelet transform (DWT) and cascade forward back-propagation neural network (CFBPNN). Extracted entropy features form discrete wavelet transform (DWT) decomposed sub-bands are feed to CFBPNN for training purpose. We use JAFFE and CK+ face database in our experimental work and observe 97.14% and 96.80% classification accuracy respectively.