The Study On Real-Time Face Tracking And Recognition Algorithm

In recent twenty years, the technique of face detection and face recognition, as one of the important research areas of computer vision and pattern recognition, attracts more and more attention. It is widely applied to numerous commercial and law areas, such as mug shots retrieval, real-time video surveillance, security system and so on. Now, recent researches focus on motionless face recognition. In order to satisfy requirements of application, this thesis puts forward the method of the motion face recognition based on the face tracking. Therefore, this thesis makes researches in three parts:The first part is the face detection. It is the incipient work and plays important role in the face recognition. This part firstly uses the algorithm of illumination compensate to correct skin color. Then it introduces the YCbCr color spaces and the algorithm of skin color region segmentation. Next it advances farther segmentation and merging algorithm. Finally it presents face verification algorithm used to detect facial features and analyses detection effect and application future combing with the experiment.The second part mainly adopts genetic particle filter (GPF) to estimate the position of the face in next frame on the basis of the result of the face detection. The traditional particle filtering needs a large number of particles and appears the phenomenon of losing the target. The crossover and mutation operations in evolutionary computation are introduced into PF to make samples move towards regions with large value of posterior density function ( PDF ) and accordingly the number of required particle is largely reduced in comparison with PF. Experiments results show that GPF presents improvements over the PF techniques regarding to robustness, accuracy and flexibility in dynamic environment. Meanwhile, GPF, which needs fewer samples, meets the demands for both precise and fast tracking in the complex background.In this part, the motion face recognition based on the face tracking uses the Hidden Markov Model (HMM). It deems that a face should be described as a whole .The HMM contains a set of hidden states that canbe imaged as a set of distinct regions of the face image of a person. The numerical characters of organs of a person are associated with each other in a state transfer model. The research indicates that the method has taunt recognition rate, moderate computational complexity and good extensibility.Corresponding algorithmic experimental results are given in the every section and indicate the effectiveness and the advantages of these algorithms.