Generalized Principal Component Analysis Of Combination Of The Face Of The Image Pre-processing And Bitmap Information

Nowadays, with the application of biometrics in personal identification, validation and safety surveillance etc., the demanding for face recognition is very urgent. Automatic human face recognition is the technology of pattern recognition, which is concerned with analyzing facial images, extracting useful information for recognition and authenticating humans.There are three important stages in face recognition system: preprocessing, feature extracting and classifier design. In this thesis, we present in-depth discussion on the first two stages.At the preprocessing stage, we proposed three preprocessing algorithms for face images, that is: Image Correction By Average Template (ICBAT), Image Correction By Color Average Template (ICBCAT) and Image Correction By Symmetry (ICBS). The fundamentals of the three algorithms are coincident: first, construct an average template; then form a coefficient through the difference between an image and the average template; finally, correct the image by this coefficient. The difference among the three algorithms lies in the difference of their average templates and the different arithmetic of their average templates. ICBAT takes the average of the training samples that belong to the same class as the average temple; ICBCAT takes the average of the R G B components of each colorized image as the average temple and ICBS takes the average of each pixel in the left-half face image and its symmetrical pixel in the right-half face image. The preprocessing algorithms proposed in this thesis can make too bright areas relatively darker, and too dark areas relatively brighter, as a whole, the gray value distribution of the whole image will tend to be even. Latter experimental results indicate that the three preprocessing algorithms can reduce bright speckles and shadows introduced into an image by lighting, thereby, they decreases faces' within-class difference and makes samples of the same class closer to the class-center, thus makes samples of different classes more discriminant. Thus the robustness of face recognition is improved under different lighting conditions.