Pose Unconstrained Face Recognition Based On Virtual Face

The rise of face recognition technology began in the 1970 s,it is one of the most widely studied and highest heat problems in the field of image processing.Face recognition has many advantages,it is more friendly,acceptable,and cost-effective.In recent years,with the development of intelligent computing and analysis technology,automatic face recognition technology is applied more and more widely.Therefore,In-depth studies of face recognition technology have great significance.With the development and breakthroughs of research in the field of artificial intelligence and machine learning,particularly the research of neural network based on depth arises again,the recognition rate of frontal face images has achieved satisfactory results.However,for images in which face pose varies,many current methods can not achieve a good result.The key to solving the pose unconstrained face recognition problem is to remove the differences between different poses,of which face pose normalizing is a commonly used method.By rebuilding profile faces to frontal faces,the MRF algorithm based on 2D model and the Blanz deformable algorithm based on 3D model can recognize profile faces.Although the idea is very intuitive and these algorithms have certain effect,but much more computing time is required,and the recognition rate is unsatisfactory.To solve the problem of multi-pose face recognition under,particularly in the recognition performance issues,this paper proposes a combination of LBP based LK-SIFT feature matching algorithm framework to identify multi-pose face.First application Lucas-Kanade algorithm training face image set different sets postures converted to its corresponding positive image of the face,and learn to get the transformation parameters,and then use these parameters to identify the library being the face image converter generates all postures shot from the side of the virtual image,followed by extraction of SIFT features of these virtual images and save,only the last stage in the recognition of images to be recognized and saved SIFT features characteristic of the virtual image matches the time of this match will feature SIFT features and the LBP features a joint.Through experimental comparison,the proposed algorithm framework for each face having a servant attitude better recognition results.