Research On Face Alignment And Matching Of Face Recognition

With the era of big data coming, personal and national information security is gradually becoming a research hotspot. And the biology recognition technology becomes researchers' favourite areas because of its security, confidentiality and convenience. In a large number of the biology recognition technology, the face recognition technology comes to the fore with its non-touchless, high efficiency, convenience, accuracy and be developed to be biometric recognition technology with highest research heat. Usually, the module of face alignment and feature extraction play an important role in the face recognition system.Aiming at this direction, a deep research is conducted in this paper. The dissertation mainly includes the following works.First of all, face recognition research history, basic technical methods are summarized,the research history and technology method of facial alignment are introduced, and the research status, application and development direction of face matching are introduced.Secondly, the face detection and image processing are discussed. The current main face detection methods are summarized and classified. Based on Harr_like feature and LBP feature, the AdaBoost face detection method is discussed in choosing the features,generating the classifier and constructing the cascade detector. By comparing the instantaneity and accuracy of the two methods, the method based on Haar_like has a better performance on describing the feature. The method based on LBP feature has a higher timeliness. After detecting, by using scale normalization and grey level transformation, the face area size is unified and the colour information is eliminated.Subsequently. the face feature alignment methods are discussed from two sides. On the one hand, based on geometry feature and the facial feature points, the Explicit Shape Regression method of the face feature alignment is introduced. The alignment experiment was conducted in different databases and the face alignment results are given. On the other hand, based on the statistical feature, face alignment based on Transform-Invariant PCA method is discussed. And then KL transformation, establishing the feature space and iteration process of the inverse composition are also discussed. Experimental verification of the method is carried out by using the standard face database, results show that the method of face alignment based on Transform-Invariant PCA shows a better performance on aligning the face and has a mutual promotion effect with face recognition.Besides, the similarity measurement issue is discussed. The usual metric method only consider the difference of a pair of sample,in order to increase the discriminative,taking into account the commonness and individuality of face samples,using combined commonness and individuality of measurement methods for face samples of similarity measure. The method is verified on different database. The result presents a satisfactory performance.In the last part of the dissertation,a face matching system is established by integrating all the algorithms.