Face Recognition Algorithm Of Image Sets Based On Rotation Invariant LBP

In the rapid development of science and technology, face recognition technology isapplied in more and more places in people’s life. The contents of its main researc h is to detecthuman face images obtained by the camera, and intercept the image of human face, and thenfacial feature extraction through an algorithm, we use this uniqueness of each person’s face tocompare their similarity by computing and then achieve recognition results. Face recognitionis divided into two categories: the first is only based on one image, called face recognitionbased on a single image, and the other is based on multiple images to recognize,called facerecognition based on image sets. The beginning to research face recognition is based on asingle image, so now most of algorithms are based on a single image, and the algorithms arerelatively mature. As the request of people to the higher the recognition rate, the direction ofresearch began to the algorithm based on image set, because it contains more comprehensiveinformation to make it up with a lot of advantages that face recognition based on single imagecan not reach.Technology of face recognition is applied widely in our daily life, such as the company’sattendance system, entrance guard system, border security system at the airport, and someinformation security protection system and so on. There are many face recognition algorithmsbased on single image, and they are more mature in life application. But the requirements inthe application of its background is demanding,and it is more sensitive to illumination,posture, shade, expression and so on. As the means of obtaining image are different, such asrandom access to the video and the image sequence at different times, different backgroundsand different positions. If we use the algorithm of single image, the recognition rate will begreatly reduced. The face recognition algorithm based on image set was supplemented itsshortcomings, because it is modeled on the basis of multiple images, it contains moreinformation about the images and it is better to express individual. At the same time it alsoeliminates the adverse factors such as light, shade. We can get the better identification, and therecognition rate will also be more stable.During graduate I read many English literature associated with face recognition, I learnlot of facial feature extraction algorithms, focused on the face recognition algorithm based onimage set. Through careful study of face recognition algorithm based on image set, I know that it can overcome the light, background, posture and other shortcomings,because itcontains a lot information of images; However, there are some disadvantages of image sets, itwill be affected if the feature is not obvious; affine package will overlap if it is too large. Inthis paper,we use multi-scale and rotation invariant LBP algorithm on the image set tooptimize model. LBP algorithm extracts the facial features of a human Prominently, andweaken other areas that is not representative.It also has rotation invariance, and reduce theinfluence of light. Gaussian multi-scale space has the effect of filter, it removes some noise ofthe picture, there is a different texture in different scale space, so it can enhance the LBPfeatures.the result of recognition rate is increased.To verify the effect of algorithm in this paper, the experiment is in the Honda/UCSDdatabase, namely Honda at the university of California, San Diego face tracking videodatabase. First of all, we use the classic Viola–Jones to detect face, in order to identify, weprocess the image to make them in unified, such as gray processing, size; Then, we do featureextraction for the detected image.In this paper we mainly use rotation invariant LBPalgorithm in Gaussian multi-scale space to extract the image feature,it can make the facialfeature more obvious, and representative; We model image set after feature extraction, webuilt affine for each image set; Finally, in this paper we use Euclidean distance to calculate thedistance of affine package, the nearest is the same face image.Through experiment many times and comparison with the original affine algorithm, thealgorithm in this paper improves the recognition rate and more stable.