Research On The Technology Of Face Recognition With Ageing Variances Based-on HMM

Face recognition has a high rapid development in recent years and been one of the most successful applications. It has been the most potential research topic of the biometric recognition for its perfect characters.However, the recognition rate falls down rapidly while recognizing different aged images of the same individual, for the face has changed greatly by time. The paper gives a comprehensive survey and analysis of the existing techniques about ageing variation and face recognition internal and overseas. By systematically analyzing of relevant algorithms, we present a novel algorithm for face recognition with ageing, added into the image reconstruction algorithm and age forecast algorithm, from two aspects of age prediction and image reconstruction. In addition, a prototype system of face recognition is designed and implemented.The highlights and main contributions of the dissertation include:(1) Standardized segmented local SVD is used to extract feature instead of whole SVD. Feature extracted methods based on SVD are studied deeply, the segmentde local SVD is used to extract the feature for features extracted can include more information of facial image and enlarge their differences. Then, these features are standardized.The experimental results have shown that the approach can improve the recognition rate.(2)An age prediction algorithm based on the Elman neural network is proposed. After analyzing the existing age prediction algorithms, the key aging features are extracted by Single Value Decompose (SVD) and the wavelet algorithm, then these features are used to train the network to derive an age prediction model. The network parameters are optimized and a better age predicted model is built to obtain the age range of the new image, using to estimate the age range of the face. The algorithm can predict the age class from facial image and the error in the predicted age of the unseen data is approximately±3 years.(3) A face restruction method based on shape and texture features is presented. Firstly, the non-linear model describing the relationship between age and key feature points in face is built based on the warping technology. Secondly, the method of filtering aging texture images by multi-orientation filters are proposed according to the idea of Gobal multi filters. The new age image of a given age is constructed by adding the according age textures to the shape-warped image selected by the built age predicted model. From the experiment result, the reconstruction images are visually realistic and gain the characteristic of the age group they belong to.(4) Based on the classify algorithm of Hidden Markov Mode, the test images are classified and recognized to testify the validity of our method proposed.(5) Based on the idea of oriented object, we design and implemt a prototype system of face recognition with ageing variances, which is divided into five modules that is image preprocess, age prediction, ageing image reconstruction, facial feature extraction and face recognition.The experiment result proves that the algorithms above are effective when test images are captured over 5 years after the subjects in question provided training images.