Study On Face Detection And Recognition

An accurate automatic personal identification is critical to a wide range of application domains such as access control, electronic commerce, and welfare benefits disbursement. Traditional personal identification methods (e.g., passwords, PEST) suffer from a number of drawbacks and are unable to satisfy the security requirement of our highly inter-connected information society. Biometrics refer to automatic identification technology of an individual based on their physiological traits such as fingerprint, face and iris or behavioral traits such as signature, speech and gait. Currently, there are many biometric techniques that are widely used. A biometric system is essentially a pattern recognition system, which makes a personal identification by establishing the authenticity of a specific physiological or behavioral characteristic of the user.Facial images are probably the most common biometric characteristic in order to make a personal identification. Face recognition is one of the most active research areas ranging from static, controlled mug shot verification to dynamic, uncontrolled face identification in a cluttered background. Face recognition is a non-intrusive technique and people generally do not have any problem in accepting face as a biometric characteristic.Main content of this thesis includes face detection in color space and face recognition based on multivariate statistics analysis. In the thesis, we try to integrate the technique of multivariable statistical, neural network and statistical physics and the theory of intelligent image processing in computer vision, and puts forward and realizes a set of method and algorithm that are valuable for practical production of airport passenger recognition system.(1) Automatic face detection in color spaceAutomatic detection and localization of human faces in two-dimensional natural and complex scene images are a difficult task. In this thesis, we propose a face detection method that integrates the results of skin-color segmentation and the relations of shape features in the candidate faces.In this thesis, we propose a non-linear color space transformation. In this non-linear color space, there is a greater contrast between skin chrominance and others such as background than that of others color space. So this will be propitious to segment face skin regions.In this thesis, we put forward an-adaptive coarse-to-fine dynamical skin color segmentation algorithm. Firstly, skin is coarse segmented by the off-line static skin color model parameters, and then the dynamical skin color model parameters, which are described using mix-gauss probability distribution, are trained by the on-line and adaptive learning process, and finally skin is segmented by this probability model, and face candidate regions can be obtained. The experimental results show that this algorithm is effective and robust to face segmentation in complex background.However, such skin color models are not effective where the spectrum of the light source varies significantly, and skin color alone is usually not sufficient to detect faces. The regions by skin color segmentation are all candidate face regions, in order to authenticate this candidate face regions is actual face, we present the method based on texture information in luminance image and PCA edge direction information for automatic detection of eyes directions and then detection nose and lips by geometrical face modeling.(2) Face recognitionThe core of face recognition is the selection of face representation and match strategy. Because of the robust of the algebra feature vector, the thesis mainly presents several this kind methods of face recognition.The eigenface method is the classical method in this kind. It analyzes the face by Principal Component Analysis (PCA), However, this method does nottake into account the noise, so it can lose much information in practical applications. In this thesis, we propose the method of face recognition based on another principal component analysis named as Probabilistic PCA (PPCA), and the PPCA is bas...