A Frontal Pose Face Detection and Classification System Based on Haar Wavelet Coefficients and Support Vector Machine

Abstract

Abstract-This paper presents an integrated face detection and classfication system for faces with frontal pose. The face detection sub-system is based on Haar wavelet coefficients and the face classification sub-system is based on support vector machines. The proposed system is trained using the VISiO multi-view face database and is tested using the commonly used test sets. Our experiments show that the proposed face detection sub-system has a 94.8% detection rate while the face classification sub-system has a 68.1% classification rate.Index Terms-Face detection, face classification, haar wavelet, support vector machine. I. INTRODUCTIONNowadays, there are a lot of applications that rely on the classification of human faces. For example, such a system can be employed as a part of a security system that allows access to a certain area only to persons that are members of a certain group. Another example is a surveillance system that can give an alert to law enforcement agencies of the presence of people that are known to belong to terrorist groups. Each of these applications relies on the integration of a face detection system and a face classification system. The face detection system is used to locate the facial area within the input image while the face classification system is used to determine to which group the detected person belongs.Ideally, a face detection system should be able to locate all the faces in a digital image, regardless of position, scale, orientation, age, expression, illumination conditions and image content [1]. There are many discriminating features that can be used to detect faces proposed in the literature. For example, one can use human skin colour detection to locate faces [1] [2]. Another approach is to use template matching methods [3]. The authors in [2] have also proposed another possible approach, by detecting the presence of objects, like nose and nostrils, which are normally present in human faces. Finally, another approach that has been proposed is the use of statistical models of the facial area and non-facial areas [4]. In our previous works [5][6], we have shown that the use of 1-dimensional Haar wavelet coefficients as a discriminating feature to detect faces can give a satisfactory result which are robust againts variations in background, illumination and subject expression.Face classification system is a machine that is used to classify people based on their face. Support Vector Machines (SVM) is one of the possible methods used to classify faces. SVM is well-known because this method utilized optimalization approach to construct a separating hyperplane as a decision surface [7]. This hyperplane is constructed in a high-dimensional feature space by using a nonlinear mapping. By the means of SVM, the face classification system is expectantly able to construct a decision surface which maximizes the margin of separation between the face images among the two groups of people.This paper presents a system that could detect the areas of an image which contain faces and subsequen...

Improved Face Detection Method via Learning Small Faces on Hard Images Based on a Deep Learning Approach

“…Although a good running speed is obtained; it does not have a satisfactory accuracy. SVMs [ 35 ] can be trained for face detection, a good example of which is the Haar wavelet. When Haar wavelets are trained on positive and negative applied examples of feature extraction, it helps to distinguish the classes; however, they faced a problem—it could not pick up the faces of various poses because they were weak, which resulted in the poor performance of the classifier and the results were indeterminate.…”