A Novel Face Emotion Recognition System for Pose and Illumination Variation Using Adaptive ANN

Abstract

Communication, whether verbal or nonverbal, is essential for completing many newlineeveryday activities and it plays an important part in life. Face emotion is an utmost newlineefficient way of non-verbal communication and it bestows an indication of sentimental newlinestatus, mentality and intent. Usually programmed face emotion recognition structure newlinecomprises of three phases: tracking of face, extracting features and classification of newlineexpressions. To build a powerful face emotion recognition structure which can yield newlinereliable results, characteristics have to be extracted using the applicable facial areas that newlinehave sound discriminatory capabilities. newlineVarious methods of programmed face emotion detection have recently been suggested, newlinehowever perpetually each of them is algorithmically costly and use time calculating the newlineentire facial picture or splits the facial picture according to a mathematic or geometric newlineheuristic for the extraction of the features. None of these are inspired by the human newlinevisual method in carrying out the same task. The human visual scheme is used in the newlineproposed research work to extract the features from face area. We maintain that the newlineprocess of analyzing and recognizing emotions might be performed more favorably, if newlinesignificant face areas are chosen for subsequent processing, like it occurs in human newlinevisual technique. Facial expressions may be a critical sign for nonverbal newlinecommunication among peoples. newlineThe task of face expression identification is transcendently perplexing for two reasons. newlineFirst reason is the unavailability of huge archive of training pictures and another reason newlineis regarding categorizing the emotions, which might be convoluted as per the state of newlinethe input image (static or dynamic). In this thesis, we focused on computer vision-based newlineschemes to identify and recognize the faces and expressions, respectively. Moreover, newlinewe also present a CNN based prototype to estimate the age and to classify the gender newlineof detected faces along with its emotions. In first, phase, we present a hybrid feature newlineextraction