Research On Face Detection Based On POCS Algorithm

With the wide use of face recognition technology, the face detection technology as one of the parts has been rising gradually and attaining much wider using spatial,such as video surveillance, sign-in statistics. Moreover, the detection rate is the significant direction of face detection research. In order to improve the accuracy, this paper is considered from two aspects, one is the quality of input image; the other one is the feasibility of detection algorithm.With the purpose of improving the quality of detecting image, this paper puts forward using both super resolution image technique and face detection technology in combination. Super-resolution image mainly surrounds POCS(Projection onto Convex Sets) and Ada-boost.while face detection technology mainly adopts the methods of template matching. And the following is the content of research innovation:1.This is the first time to apply super-resolution image technique based on POCS to face detection technology so that we can achieve the aim of improving detection effect by the combination of the two.2.On account of improving traditional POCS algorithm, we come up with eight-direction operator as the method of pretreatment. We can conduct super-resolution restoration according to low-resolution image; and do quality evaluation of the result. Taking measures of super-resolution restoration evaluation on the basis of SSIM-NCCDFT, proving the improvement of revised POCS algorithm of structure similarity and Fourier spectrum in image, compared with traditional one by experiments.3.To conduct face detection on the basis of super-resolution in image detection,the arithmetic adopts traditional template matching. However, we need to improve,because of the imperfection and shortages of the depth of detected face spinning and low accuracy. Thus based on face detection method of half-face model, we need to improve. First, we will select roughly with the method above; and then detect faceagain by Ada-boost arithmetic. With the use of ORL face database as template matching, we can finally detect the depth and accuracy of face spinning by experiments and find that the combination of matching detection of half-face model and Ada-boost arithmetic makes the face detection more flexible and accurate.