Age Estimation Algorithm Based On Facial Images

The facial image contains many important personal information, such as gender, age, race, etc. In recent years, researchers paid much attention on age estimation of facial images, and this problem has become one of the emerging topic in the field of computer vision. This thesis mainly focuses on facial images to proceed age estimation methods, especially on the feature extraction and age estimation. By steps of the face image preprocessing, feature extraction and classification of age estimation procedures, we implement the entire age estimation task, the specific work includes the following aspects:(1) Facial images preprocessing for age estimation Because the source of the facial images in the database are not uniform, there is inconsistency between the head pose and illumination etc., therefore, before the feature extraction, we need preprocessing the input images to highlight the face region, so as to get the normalized images. In order to get better age estimation effect, this thesis uses the method of face detection to locate the interesting face region, the method of image enhancement to highlight the human face details, and normalize all the samples to the same size.(2) Facial images feature extraction for the age estimation For the expression of information about the face age, we consider that the shape information and texture information fusion, which is much able to describe the face age information. In this thesis, we employed the idea of Canonical Correlation Analysis(CCA), and proposed a novel age estimation method based on feature fusion. More specifically, the method will firstly extract two sets of feature vectors for the same pattern exploiting Active Appearance Model (AAM) and Local Binary Pattern (LBP), and building a correlation criterion function for the extracted two sets of feature vector. Then, creating two sets canonical projective vectors with the proposed strategy. It's a method which applying canonical correlation discriminant features extracted by specified feature fusion strategy as the facial features. For the change of facial age, provides beneficial complement from the shape and the texture.(3) Age estimation method based on Extreme Learning Machine (ELM).For the establishment of age estimation model, existing methods cannot solve the problems such as long training time and model parameters, but extreme learning machine (ELM) can well avoid these difficulties. In this thesis, the ELM theory has been studied for the age estimation. We. proposed use the ELM to build the age estimation model and we first introduce ELM to the field of age estimation. With its fast learning speed and good generalization performance, the parameters can be directly determined, combining with the thought of hierarchy, we built a two kinds of classification which can use to estimate age. First roughly estimate the sample's age, then do regression in their respective categories by Extreme Learning Machine regression, to get a strong generalization ability of classifier with the combination of shape and texture features. The established classifier is used to estimate age, get specific estimate age value.The experimental results show that the proposed method achieves smaller mean absolute error (MAE) and better estimation accuracy, and the final MAE is within 5 years.