Face localization by closed loop discriminator estimation and improved detection using contemporary feature extraction techniques

The present paper describes a novel adaptive cross correlation technique of face recognition using closed loop discriminator estimation for face detection. All possible variations in the human face can be obtained by scaling and three types of head roll in different planes. We show here that face recognition system comprises face detection and face verification. Feature selection schemes like eigenfaces, local binary pattern (LBP) and speeded up robust features are implemented to extract features with high discriminative power, thereby performing face verification. Experiments revealed that this method strikes a balance between accurate face recognition and identification with sufficient speed of convergence. The algorithm proposed by this paper allows tracking of face against a wide scale range with sufficient immunity against like camera vibrations, sensor errors, illumination level fluctuation etc.

I. Introduction

One of the most important tasks of image processing has been face recognition courtesy the requirements of enforcement of security and widespread commercial applications. Among the diverse multimedia contents available for identification purpose, face recognition is important because it is nonintrusive, can be used without cooperation of the subject [1] and due to availability of maximum number of prominent feature vectors. Automatic tracking and identification of facial features is an active area of research. Face detection module requires face localization and face alignment; scaling and rotation are two types of alignment operations performed on image set. For multi-face image or identity authentication stricter verification techniques involving feature extraction and statistical methods are implemented. Reduction in core dimensionality of feature matrix [2] by suitable algorithm ensures faster convergence but not at the cost of loss in uniqueness of identity. Development in the field of image processing hardware in terms of cost and miniaturization has boosted the face recognition software market notably in areas such as speed and accuracy.