Face Detection And Recognition Method

Modern society's fast development has a higher request on the safety, veracity and applicability of identification. However, the traditional distinguish ways such as password, credit card etc can not meet the requirements of modern society as well as increasingly fail to meet modern science and technology development and social progress. Biometric technology makes it all possible. Biometrics refers to automatic individual identification technology based on their physiological traits such as fingerprint, face and iris or behavioral traits such as signature, speech and gait. In these technologies, the face identification develops fastest. Relatively speaking, face recognition is a more direct, more convenient, more friendly, and easy-acceptable identification method. The face identification technology gives computer the ability to identify according to the face, which has a wide range of using. With the help of computer vision, photoshop and neural networks, face identification technology has been an advancing front and hot issue in the world.Based on analyzing and summarizing the relevant research fruits at home and abroad. This thesis has done experiments and researches on face detection and recognition. It has improved the identification methods and got some good results in identification time and accuracy.The main research in this thesis focus on the following two aspects: face detection based on the static color-image and face recognition using Artificial Neural Networks (ANNs) based on Error-Correcting Output Codes (ECOC). In the face detection, firstly pick up the skin region (candidate region) where may have human faces by using skin detection; And then image preprocess is used in these regions in order to make the area clear; At last, the face detection model based on Artificial Neural Networks is used to detect the faces (number, size, location, etc.). This model is much faster compared with the traditional face detection method. In the face recognition, feature selection or feature extraction is necessary because of the huge data. In this thesis, the Discrete Cosine Transform (DCT) is used. Generally speaking, face recognition is a classification problem with multi-classes, so the Error-Correcting Output Codes is used to make face recognition becomes a classification problem with only two classes in order to make the problem easier. In this two-classes problem, a Hybrid Artificial Neural Network (HANN) model is proposed. Face features selected by DCT are put into this model, and every ANN will have an output between -1 and 1, then these values are put into the compare system to get the results. Empirical results indicate that the proposed framework is efficient for face recognition.