Facial Expression Recognition Using Modified Hidden Markov Model

Abstract

This dissertation work introduces the novel framework on Facial Expression newlineRecognition (FER) using Modified Hidden Markov Model (MHMM). Two layer newlinearchitecture of Modified HMM is applied to map the most similar facial features. Two newlinelayer modification of HMM comprises of Upper and Bottom layer. The bottom layer is newlineused to identify atomic expressions formed by mouth, nose, eyes separately and the newlineupper layer formed by combinations of atomic expressions such as joy, fear etc. From newlinethe combination of various feature extraction or face descriptor with Modified HMM, it newlinecan be used efficiently and robustly to recognise various facial expressions. The results newlineare obtained using a public dataset called Japanese Female Facial Expression (JAFFE) newlinewith 13 subjects in which one image is used to test and remaining images are used to newlinetrain the Modified HMM classifier. It is also shown that how two layered extensions of newlinenormal HMM is used to get a high recognition rate of 85%. newlineThe combination of various features extraction techniques with two layered extensions newlineof HMM, which forms a better and flexible technique to identify facial expressions. It is newlinealso shown how the layered extension is used to get better accuracy. newlineThe performance of Modified HMM is compared with other previous methods and newlineexperiments are also conducted in MATLAB2013 environment. Finally, the results of newlineour framework are carried out and images are taken using JAFFE. We have also proved newlinethat our new framework is superior from other previous frameworks using accuracy, newlineprocessing time, recognition results, ROC, errors, and found the accuracy of 85%, newlinewhich is better than previous frameworks. newlineIn the future, Modified HMM will be used with other methods such as Complex newlineMoments, Rotational Moments and Zernike moments, and compare it with other newlineprevious methods and try to enhance the results. In the future, we will also incorporate newlinecontempt facial expressions, spontaneous expressions, low-resolution images, and realtime images in our framework. It can be also applied in