Research On Face Recognition Algorithm With Age Robustness

With the continuous development of society and technology, secure identity authentication is becoming an urgent need. Biometric authentication has been developing rapidly in recent years and gradually replacing the traditional way. Among biometric technologies, face recognition is widely used in identity authentication, police investigators and daily attendance, because of its convenience, safety and university. However, the images in the database exhibit a certain age variability with the test data, face recognition rate decline caused by aging is becoming a challenging problem. In this thesis, we studied the face recognition technology under the changes of age, and added age estimation module and face reconstruction module to the existing identification system framework. By this mean, the recognition efficiency is improved significantly.The major contributions of this thesis could be summarized as following:(1) We proposed an age estimation method which combined the LGBP features and the support vector machine regression. First we conducted the Gabor wavelet transformation on face image, and coding the amplitude maps by LBP operator to get the LGBP features. Then the PCA method is used to reduce the dimensions of the feature vectors. Finally, support vector machine (SVM) regression algorithm is used for training of age estimation function which matchs texture characteristics and age tags, the algorithm achieves good effects on age estimation.(2) We proposed a face image reconstruction method based on image deformation technique and texture superposition technology. First, facial features are calibrated according to AAM algorithm. Then the improved texture transformation method is used to deal with the image after deformation. Therefore, we can get the reconstructed face image which contains more age information. As demonstrated by experiments, the proposed face reconstruction model is very effective in simulating the aging of faces.(3) We studied the face recognition technology base on the GOP model. In this thesis, the GOP model is used for face description and the SVM is used for recognition. By combining the age estimation and face image reconstruction method mentioned above, we designed a new face recognition system framework and achieve good recognition results.