Research On Face Recognition Algorithms Of Image Sets Based On LBP

Face recognition belongs to the biometric identification technology, which is mainlyabout the characteristics of the human face. It uses the characteristics of the face to achievethe purpose of distinction the other individuals, by using the face recognition technology.Currently, the face recognition technology has been widely used in various fields of life.Compared to other types of biometrics, face recognition has broad application prospects,because it has the characteristics of the simple and intuitive, not easy to copy, high safety,easy operation, and so on.Currently, in the face recognition technology, the largest study of face recognitiontechnology based on single image, and its theory and technology have been more mature. Butthe recognition algorithm based on single image has some shortcomings: the light of thesurrounding environmentally, the change of people’s posture and expression, the human faceof the partial occlusion, a matter of age; the quality of the captured image etc., these factorswill affect the final recognition rate.With the rapid proliferation of video surveillance, and the development of video capturetechnology, sometimes we captured image is not ideal, and is not up to the requirement of theface recognition technology based on single image. Then, the face recognition method basedon image set to get more and more attention. This method is no longer to provide a singleimage of unknown class, but to provide images of the same object in a group of unknownclass. This set of images collected by video or through the different periods, scene, and thelight to get, these images are composed of a collection to represent an individual, and theindividual conduct of training and recognition. Because individual representation based onimage set, consists more information, so this can more effectively represent the individual,and easy to eliminate some negative factors, such as lighting, posture, facial occlusion, etc., sothat this method makes the recognition rate more stable.Compared with the face recognition based on a single image, the face recognition basedon image sets has better robustness, but also has better performance. The emergence of facerecognition algorithm based on image sets, making the face recognition technology hasimproved, and also make the application of face recognition more widely.By reading the literature of the face recognition method based on image set, this paperfocuses on the research of face recognition algorithm based on the affine hull, and the LBP(Local Binary Pattern) algorithm is applied to the image feature extraction.We use the Viola-Jones face detection algorithm to extract the images (grayscale image)from the video stream, conduct of face detection, localization, extraction for these images, set the obtained face images with the uniform size, and build the face database through theextracted images for our experiments. Using the Local Binary Pattern (LBP) featureextraction algorithm for the face images which obtained by the Viola-Jones face detectionalgorithm, and keeping the effective information of face image. Based on this, we use theaffine hull method to establish the model of the image set which is conducted of the imagefeature extraction. And the model is optimized so that there is no intersection between themodels. We can measure the similarity between the two models by the Euclidean distancemethod, which is that, through calculating the distance between the test individual model andall the training individual model, we can get the shortest distance data, and put the unknownindividual classified in the training individual which is correspond to the shortest distancedata.In order to get the recognition rate of the face recognition algorithms of image sets basedon LBP, the algorithm which we conducted experiments on the Honda/UCSD video database,and compared with the original recognition algorithm of image sets. The experimental resultsshow that the face recognition algorithm of image sets based on LBP has the betterrecognition rate.