Font Size: a A A

Multi-Pose Virtual Face Recognition Algorithm Based On 3D Model

Posted on:2017-05-03Degree:MasterType:Thesis
Country:ChinaCandidate:R LiuFull Text:PDF
GTID:2348330503465424Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Face recognition technology is a very important identity identification technology based on biological characteristics. It uses the computer to analyze the human face image and extract the effective feature information, and then the matching and classification of objects can be achieved based on the feature information. In the practical life, face recognition technology is more convenient, friendly and so on, so it is widely used in many fields, such as government, commerce and so on. After several decades of development, it covers a number of disciplines, such as mathematics, computer science, psychology and so on. It has important guiding significance for the development of various disciplines. At present, face recognition technology has become the forward subject in the field of machine vision and pattern recognition, and it is of great significance to deeply study the technology of face recognition.At present, many face recognition algorithms can achieve good results in ideal environment, but the influence of environmental changes on the recognition results is even greater than the differences between human face categories in practical application. In practical application, the collection standard frontal face image is difficult, so the pose change has brought great challenges to face recognition, and it has become one of the most classical problems in face recognition, pattern recognition and machine vision. Aiming at the problem of multi-pose face recognition, this paper mainly studies the multi-pose virtual human face recognition based on 3D model. The core idea is that it turns all the frontal images in the face database into multi-pose virtual views, and then the features of virtual face images and test image are extracted respectively, and then the classifier is used to match the features and output the recognition results.First of all, this paper introduces the background and significance of face recognition, and summarizes the research status of existing face recognition algorithms and the main challenges of face recognition. And the practical value and theoretical significance of multi-pose face recognition are analyzed. The main research results and methods of multi-pose face recognition in recent decades are introduced in detail based on two aspects of pose correction and generation of multi-pose virtual views, and their respective advantages and disadvantages and performance are described and compared in detail.Then, a virtual face generation algorithm based on 3D model is proposed in this paper. At first step, the relationship of face deflection in 3D space and two dimensional is introduced, and derive the relation function between face deflection and angle in two dimensional. And then, multi-pose virtual face images are synthesized by the relation function. In addition, this paper also introduces a virtual image generation algorithm based on polynomial function fitting, and compares and analyzes the experimental results of this algorithm and the algorithm of this paper.Finally, the multi-pose face generation algorithm based on 3D model is applied to face recognition based on single view. Firstly, the deflection angle of test sample is obtained by the Ada Boost algorithm and the integral projection function. Then, the multi-pose virtual face of training samples are generated by the multi-pose face generation algorithm based on 3D model, and the training samples are increased. Finally, the classification method based on Fisherface is used to achieve the multi-pose face recognition.
Keywords/Search Tags:Multi-pose, Virtual Face, Face Recognition, 3D Model, Two-dimensional
PDF Full Text Request
Related items