Automatic Human Age Estimation Based On Layered And Weighted Support Vector Machine Model

Human face image reflects important age information, face age estimation has important research significance and application value. To explore face age estimation problems in depth will improve the research of pattern recognition, artificial intelligence, robotics and etc. Currently, the demanding of automatically updating face database in public safety filed, collecting the information on customers of different ages purchasing goods in commerce filed, and simulating the character aging progress in film and television production filed has become increasingly imperative.Thesis firstly proposes Layered and Weighted Support Vector Machine age estimation model by the research of face age estimation at home and abroad. Age estimation models mainly composed of facial features extraction, model training and age estimation.Currently, some of the age estimation model extracts facial features from the whole image, these features contain a lot of redundant information. Thesis adopts AAM combined appearance model to extract features from facial shape region, these features integrate the human face shape and skin texture information, and abandon a lot of redundant information, so it can better fully reflect the age characteristics from human face.In order to achieve the aim of automatic face age estimation, which need to find out the human face from image. At present, the result of face fitting algorithms for locating facial landmark points is unsatisfactory. Thesis proposes a new face fitting algorithm named LTC_RSIC_AAM based on local texture constraint from existing algorithm. Compared to current fitting algorithms, LTC_RSIC_AAM algorithm not only locates landmark points in facial contour more accurately, but also fits the shape of face more accurately.Because support vector machine shows perfect performance in pattern classification, so thesis uses SVM age estimation models to classify and predict the features data. For the first time thesis puts the layered and weighted mechanism into support vector machine. As people of different ages have different age progress laws, thesis divides features data by age and trains sub-models, and then predicts the age of features data by different sub-models. In order to solve the classification bias problem caused by the unbalanced number of two data samples, by weighting the penalty parameters which further improves the accuracy of age estimation. The experiment results show that the proposed automatic face age estimation method based on Layered and Weighted Support Vector Machine Model achieves the result of age estimation much better than the most of popular methods at present.Experiment shows that the proposed automatic face age estimation method based on Layered and Weighted Support Vector Machine Model got much better results than the present known methods in the field of age estimation. The mean absolute error of age estimation reduced to 4 years, reaches 3.85 years.