International Conference on Technology, Engineering and Science (IConTES 26-27 October 2017). Antalya, Turkey ANALYSIS OF FACIAL CHARACTERISTICS Fatma GÖNGÖR Adana Science and Technology University, Electrical and Electronic Engineering Department, Adana, Turkey ftmgongor@gmail.com Önder TUTSOY Adana Science and Technology University, Electrical and Electronic Engineering Department, Adana, Turkey otutsoy@adanabtu.edu.tr Corresponding author: Onder Tutsoy, otutsoy@adanabtu.edu.tr ABSTRACT: Recently, physiognomic evaluation of human face has been widely considered in computer based visual applications. This paper presents a facial character analysis algorithm and its application to a person. This algorithm has 3- stages: Initially, face is detected from images with Viola-Jones algorithm and then crucial facial distance measurements are measured with Geometric based facial distance measurement technique. Finally, measured facial distances are evaluated with Physiognomy science to interpret characteristic properties of the person based on nose-forehead, mouth-chin and eyes-cheeks facial measurements. The simulation results show that performed facial character analysis reveals important information about the character of the experimented person. Keywords: Facial character analysis, Geometric based facial distance measurement, physiognomy, Viola-Jones algorithm. Acknowledgement: We would like to thanks to TUBITAK as this work is supported by TUBITAK with project 215E047. 1 INTRODUCTION Physiognomy deals with character analyzes by interpreting physical properties of the body and face. Especially examining the face features reveal crucial information about the character of the people. Facial character analysis based on facial features are widely used in a variety of areas such as fraud detection and criminal judging applications. The proposed algorithm for facial character analysis consists of three key stages: Human face detection, facial distance measurements and physiognomy based interpretation. Face detection is the first stage of the facial character analysis. In the literature, a number of promising algorithms have been proposed. Initial work of Viola-Jones aimed to developing an algorithm to only detect frontal faces [1-2]. This forms a basis for the succeeding Viola-Jones algorithms which are able to detect faces from various angles and profiles. Rotation invariant neural network based 3D face detection algorithm proposed and experimented on a wider range of face images than Viola- Jones algorithm [3]. Even though this work shared a number of commonalities with Viola-Jones algorithm, its face detection speed and accuracy were lower. It was stated that the Viola-Jones algorithm required less detection time compare to these algorithms although the amount of scanning time was same [3]. While the key reason for Viola-Jones algorithm being faster is the implementation of a boosting algorithm called Ada-Boost, the Cascade classifier improves the correct face detection rate of Viola-Jones algorithm [12]. These are the motivations behind selecting Viola-Jones algorithm for face detection in this work. After detecting the face, the second stage of the facial character analysis is to obtain facial distance measurements. In this term, one of the most effective technique which is Geometric based facial distance measurement technique preferred to use for this research. Recently, Valstar et al. have stated that Geometric based techniques provide higher accuracy in facial measurements than appearance based techniques [14]. One of the earliest Geometric based techniques was proposed by Kanade and tested on 16 facial images which showed that the accuracy of the facial measurements was 75% [15]. Brunelli and Poggio modified Kanade’s Geometric based technique and examined it on 35 facial images which yielded 90% recognition accuracy [16]. Finally, physiognomy based interpretation of the measured facial distances is performed to comment on the character of the person. One of the earliest physiognomic interpretation of human characteristics was studied by Bell and his collogues [17]. In