1 Real Time 3D Facial Emotion Classification using a Digital Signal PIC Microcontroller Ahmed FNAIECH 1 , Sami BOUZAIANE 2 , Mounir SAYADI 1 , Nicolas LOUIS 3 and Philippe GORCE 3 1 Université de Tunis, Labo SIME, ENSIT,,Tunisia 2 Naval Academy Menzel Bourguiba, 7050, Bizerte, Tunisia 3 Labo HANDIBIO, University of Toulon, Toulon, France E-mails: ahmedfnaiech@hotmail.com, samibouzaiane@yahoo.fr, mounir.sayadi@esstt.rnu.tn, nicolas.louis83@gmail.com,gorce@univ-tln.fr Abstract:The human face is viewed as the mirror that reflects the inner feelings of the person and allows us to detect the needs of every person and study his behaviour and requirements. We can also identify the tastes of each person and predict his reaction through facial features. Indeed, the detection, classification, characterization of the face are research areas that have received considerable interest in the recent twenty years. However, most published works in this field study the emotions for a person in an upright posture, which means that the person must perfectly face the camera. The innovation in the present work is to improve the detection of emotions in the case of different face orientations (to the right, left, up and down). Eight-teen feature points that perfectly characterize the human face are firstly calculated. An optimization step is proposed by extracting a set of optimal distances between the facial points as a new set of optimized emotion descriptors. For this reason, we have used a statistical characterization criterion based on the ratio of the intra- class variance. An emotion classification experiment is carried out using a multilayer neural network implemented with a digital signal processing microcontroller. Key words:emotion detection, features optimization, characterization degree, neural network classification. 1. Introduction Facial emotion recognition is a rich field which is actually undergoing considerable development and vertiginous expansion, since it touches to the cognitive state of a human being, his behaviour and reactions. In fact, many works were interested in facial emotion detection and classification in the 2D field and the 3D field, but most of these have developed facial recognition algorithms or-well sensing facial emotions under standard conditions i.e. detect emotion for a person in an upright posture. In other words, the person must perfectly face the camera which may be considered as a hindering drawback. It is often useful to know what facial expressions correspond to each specific emotion. The classification of facial emotions allows us to extract important information of the human face and well describe the cognitive state of a person. The shapes and facial movement 2D view gives us important information about the expressions of the human face. Such information may vary with lighting conditions, which has imposed serious obstacles for multiple facial analysis 2D. In this context, the exploration and exploitation of geometric information 3D is necessary for solving various research issues for facial recognition. In [1], A. Metallinou et al., have proposed an audio- visual emotion classification using hidden Markov models (HMM). In [3], A. Dhall et al., have proposed an emotion recognition method using the pyramid of histogram of gradients (PHOG) and local phase quantization (LPQ) features, and they used SVM for classification. I. Mpiperis et al. [17] have tested an approach by generating a 3D model of the face which becomes deformed elastically to correspond to facial surface; these points were then used as a base for classification. H. Tang and T. R. Niese et al. [19, 20] have proposed a method based on pattern recognition techniques for the extraction of image features. In their work, the camera models were applied with an initial stage of record in which the face of the person was automatically built starting from stereo images. The measurements of the geometrical characteristics are calculated and standardized by using photogrammetry techniques. J. Wang, et al. [21] have studied the utility of the geometrical shapes of the face to represent and to recognize the facial expressions in 3D. They also proposed a new approach to extract the primitive characteristics from the facial emotions. Then, they applied the characteristic distribution in order to classify the expressions of the face. A. Maalej et al. [22] have used an approach based on the form analysis of local “patches” extracts starting from 3D model faces. Quantitative similarity measures were then obtained and used as input parameters for the algorithms to multi- class classification. M. Lyons et al. [23] have presented a method where the facial expressions are coded by using the multi-orientation and multi-resolution filter of Gabor which are ordered and topographically aligned roughly with the face. Their results showed that it is possible to classify the facial expressions with the filter of Gabor. The novelty in the present work is to improve the detection of emotions in informal conditions. In fact, we consider the face reactions from various angles with a gradual variation of the divert angle ranging from 0 degree up to 30 degrees. Each face is characterized by a set of 18 specific points which allow the recognition of the human face by the calculation of distances between these points. Seven different emotions are taken account in this study such: joy, fear, sadness, surprise, anger and 2018 IEEE International Conference on Image Processing, Applications and Systems (IPAS) 285 978-1-7281-0247-4/18/$31.00 ©2018 IEEE