In the past several decades, significant progress has been made in the field of computer vision and machine learning. The learning algorithm has been changed from rule-based to statistics-based. What’s more, as the rapid increase in computer hard-ware performance and the rapid decrease in the computer hardware cost, face recogni-tion are being used in more and more applications. For example, access control sys-tems, the application of a variety of mobile terminals, smart monitoring and so on. Based on the face recognition technology, we did intensive research in the field of al-gorithm design, algorithm optimization, optimization for specific platforms.Firstly, the key steps of the Viola-Jones face detector, including the Haar-feature-based weak classifier, the AdaBoost-based strong classifier and the at-tentional cascade are studied. And we optimized the algorithm in detail. The experi-mental results show the great improvement in the efficiency, and a slight increasement in the detection rate.Secondly, a framework for eye detection is presented,which divide eye detection into two steps, including coarse detection and fine detection. The coarse detection can be implemented by Viola-Jones detector. The fine detection can be implemented by either Viola-Jones-based verification or mean shift method. These two different me-thods are suitable for different scenarios respectively.Thirdly, after deep-study of the affine transformation, a novel method for face normalization based on correlation between adjacent pixels is proposed. The experi-mental results show significant improvement in the efficiency of the normalization.Fourthly, to speed up the proposed Gabor-Orientation-Histogram-based face ve-rification method, FFTW is used in the implementation. Experimental results show that the FFTW-based method increase the processing speed to18%of the original time.Fifthly, the implementation of the face recognition in the C++programming language and the C programming language is completed. And we optimize the calcu-lus from float-point to fixed-point. In addition, by the use of Android NDK, the face recognition system can be invoked by Android applications. |