Research On Part-based Face Recognition

Face recognition is one of the most prominent areas in computer vision for several decades.Numerous face recognition methods have been proposed.Current systems can do fairly accurate recognition under constrained scenarios using these face recognition methods.It has attracted much attention due to its potential application values. Nowadays,face recognition has great value in public safety,intelligence surveillance, identification,E-commerce,multimedia,digital entertainment and so on.Requirements of the practicality of the face recognition is increased along with the wider use.Also,most of the current face recognition methods can do fairly accurate recognition in fixed environment.But the fixed environment constrain is not suitable for most of the application.This paper discuss the problems mentioned above,the following contributions are made:●A statistical learning based face decomposition method is proposed.In order to overcome different situation of human faces,such as shading and emotion. Some scholars proposed part-based or component based face recognition methods. But the decomposition of human faces are all guided under the scholars' experiences.This paper proposed a statistical learning based face decomposition methods.Our method decomposes the human face more scientifically,and reached a higher performance.In the learning process of the classifiers,the decomposition is also learned.The contribution of each part is given.Through the analysis,the decomposition is guided under each part's contribution.In the basis of the decomposition method,we proposed a part-based face recognition method.The proposed method has a 4.53%performance improvement than the original face recognition method.●An universal biometric image quality assessment method is proposed.This paper gives out the definition of Biometric Image Quality and image distance function and so on.Through the definitions,the Biometric image recognition is simplified. The unavoidable overfitting phenomenon is used to form an universal biometric image quality assessment.The overfitting phenomenon makes that the test result is better in the training set than the unknown test set.Based on that, the difference between the unknown image and the training set images can be used to assess its quality.●The biometric image quality is used in multi-classifiers fusion.The image quality of each part is used to modify its weights to reach a better performance. Each part's image quality is used to evaluates the confidence level of each part's recognition result.According to this confidence level,each part's result is dynamic weighted.This method can dynamically change each part's weight,to get more information in the higher quality parts,while getting less information in the lower quality parts.Thus,reduce the bad impact of the lower quality parts. Our experiments shows that our method can reach a better performance and be more robust to facial deformation.In the same condition,our proposed method has a 5%performance improvement than the weighed sum method,and 5.6% improvement the holistic method.