Research On Key Technologies On Human Face Recognition

In recent years,artificial intelligence and internet technology have developed rapidly.Thanks to them,several smart cities of modern society have taken initial shape,where human's recognition and identification are one of the most important parts.Face recognition technology has been regarded as a widely favored technology of internet security.Recently,face technology has made great breakthrough,improving the accuracy of recognition and identification significantly,even outperforming human's performance,which lays solid foundation for the practical application of this technology.Face technology is now applied to access control,attendance management,online payment of bank,and so on.And people are also trying to apply this technology to public security and finance security.However,applications which are more and more widely used,especially involving the safety of people and public property,make more demands on accuracy and reliability,and thus bring new challenge for face technology.For example,how to cope with the deficiency of face recognition and identification in complicated wild environment? How to do recognition using low-quality face images without decreasing the accuracy? How to reduce the impact of light variance,face pose,expression,age?In order to deal with the challenges mentioned above,this thesis proposes a face verification algorithms and two age-invariant face recognition algorithms based on metric learning.The main innovations of this thesis are as follows:1.A novel nonlinear face verification model based on sigmoid function is proposed.This thesis considers the face verification problem,which is to determine whether two face images belong to the same subject or not.Although many research efforts have focused on this problem,it still remains a challenging problem due to large intra-personal variations in imaging conditions,such as illumination,pose,expression and occlusion.The proposed method is based on the idea that we would like the similarity between positive pairs larger than negative pairs,and to obtain a similarity estimation of two images.The decision function is constructed by incorporating bilinear similarity and Mahalanobis distance to the sigmoid function.This function makes the proposed method discriminative for inter-personal differences and invariant to intra-personal variations such as pose/lighting/expression.What's more,the objective function is convex,which guarantees global minimum.The proposed method belongs to nonlinear metric which is more robust to handle heterogeneous data than linear metric.This thesis evaluates the proposed verification method on the challenging Labeled Faces in the Wild(LFW)database.Experimental results demonstrate that the proposed method outperforms state-of-the-art methods such as Joint Bayesian under the unrestricted setting of LFW.2.A new age-invariant face recognition method inspired by eigen-face is proposed.From the observation of the age-invariant face recognition method CARC,we find that although this method makes great improvement,it still has some drawbacks.First,the reference set is extremely large,since it has to cover the diversity among race,gender and so on.Second,due to the expensive computational cost,it can't make full use of all training individuals.Last,in sparse coding,CARC only adds locality constraint to ensure the smoothness,but the global distribution may be changed.To tackle these problems,this thesis proposes an eigenaging reference coding method(EARC).Inspired by eigen-face,EARC uses PCA to select some eigen components of face features.Instead of tracing the aging process for specific individuals,the proposed method traces the aging process of these eigen-face components as eigen-aging reference.The method not only improves the accuracy upon CARC,but also reduces the computation greatly.3.Another new age-invariant face recognition method based on robust feature mapping is proposed,which is called feature mapping and encoding method(FMEM).Large age range is a serious obstacle for automatic face recognition.Although many promising results have been reported,it still remains a challenging problem due to significant intra-class variations caused by the aging process.This thesis aims to remove age-related information from the original feature and obtains an expressive age-invariant feature such that it is robust to intra-personal variance and discriminative to different subjects.To achieve this goal,FMEM maps the original feature to a new space in which the feature is robust to noise and large intra-personal variations caused by aging face images,and then encodes the mapped feature into an age-invariant representation.After mapping and encoding,we obtain the robust and discriminative feature for the specific purpose of age-invariant face recognition.To show the effectiveness and generalizability of the proposed method,we conduct experiments on two well-known public domain databases for age-invariant face recognition: Cross-Age Celebrity Dataset(CACD,the largest publicly available cross-age face dataset)and MORPH dataset.Experiments show that the proposed method achieves state-of-the-art results on these two challenging datasets.