Research On Embedded Mobile Terminal Face Recognition Method Based On Cloud Computing

With the coming of information era and the strengthening of national defense construction, biometrics has become the focus and hot research technology in pattern recognition field. As one of the most typical biometric identification technology, face recognition has been widely concerned and expected. The most important characteristics of human face recognition technology are effective, convenient and noninvasive, the technology has a very good development potential and application value. Though face recognition technology has made remarkable achievements in various aspects, but it is still not enough to be fully extended to the practical application of the society. Mainly because of the research of face recognition technology is still at the experimental stage, the recognition efficiency did not fully meet the requirements of real practical application. This thesis mainly studies the feature extraction phase methods:Two Dimensional Principal Component Analysis, Non-Iteration Bilateral Projection Based 2DPCA (NIB2DPCA), Multi-Directional 2DPCA (MD2DPCA), which are mainly based on the method Principal Component Analysis. On the basis of 2DPCA and its improved method, this thesis proposes a new image feature extraction method and applies in face recognition. The main research work of this thesis has the following four parts:(1)Analyses and compares the PC A method and its typical improved methods, Summarizing the principle and realization process of each method, the advantages and disadvantages of the method while applying to face recognition are also summed up.(2)Proposes the method Multi-Directional Non-Iteration Bilateral Projection Based Two Dimensional Principal Component Analysis (MDNIB2DPCA). MDNIB2DPCA can effectively extract image features from multiple directions, retain important image information and enhance the image feature identification ability; compared to 2DPCA method, N1B2DPCA can extract feature more rapidly. Therefore, combining the advantages of the two methods, MDNIB2DPCA method is presented, and applied to gray-scale face image recognition. A large number of experiments on FERET and AR two standard gray face databases show that:this method can retain more abundant face image structural information, the identification ability is stronger and obtain a more than 2 percents higher recognition rate than methods in reference [58] and reference [59]. Compared with the MD2DPCA method, the feature extraction rate of this method increased more than 20%.(3)Combines the proposed method MDNIB2DPCA in the second item with fast color image feature extraction, then implements the color face recognition method based of MDNIB2DPCA. This method combines rich feature extraction advantage of MDNIB2DPCA and channel decomposition advantage of fast color image feature extraction, decomposed the rotated and resized color image into R, G and B channels, then operates each channel with directional NIB2DPCA, feature pre-extraction and second feature extraction, and finally use the matching score level fusion method to classify and recognize images. That a large number experiments on the two standard color face database FEI and CVL show that:this method obtains an at least one percent higher recognition rate than fast color image feature extraction method.(4)Designs and implements a face recognition test system based on cloud computing platform and mobile terminal, and uses the method proposed in this thesis to realize the image feature extraction.