Age Estimation Based On Face Images

With the development of computer vision technology and machine learning,face-related detection and recognition technology is becoming more and more mature,and face-related technology has a great impact on human life,so people have more and more attention to face-related technology.Human face images contain important information about individuals,such as human identity,gender,expression,age,color,race,etc.,which play an important role in life.Age estimation based on face images is a very important research part in the field of computer vision and pattern recognition.Facial age depends on many factors,such as beard,wrinkle,skin color,hairstyle,scar,and shooting angle,illumination,expression,wearing ornaments,etc.And one's age-related features are also slightly different for behavior patterns,lifestyles,hobbies and different age groups.Therefore,age estimation is a challenging problem in the field of computer vision.In this thesis,we study facial age feature detection and estimation based on face images,and put forward two algorithms for age estimation.(1)Age estimation based on age grouping and local facial features fusion.Given a face image,texture features based on LBP,block LBP,HOG,and GOP are extracted from the whole face and its four local facial parts including left eye,right eye,nose and mouth.With global and local texture feature description,face image is firstly classified into age groups via linear SVC,where facial age value will be estimated based linear SVR.Experiments on MORPH dataset show that the average absolute error for this method equals 2.28 years.(2)Age estimation via supervised multi-channel convolution neural networks based global-local features.First the whole face image and its four facial parts including left eye,right eye,nose and mouth are used for supervised global-local facial age features detection via multi-channel convolution neural networks.Then the global-local facial age features are combined and fed to a BPNN-based age regression model for facial age-value estimation.Experimental results on MORPH dataset demonstrate the effectiveness of the given method.