Font Size: a A A

Research And Application Of Multi-pose Face Systhesis Based On AAM

Posted on:2016-02-14Degree:MasterType:Thesis
Country:ChinaCandidate:L ZuoFull Text:PDF
GTID:2308330482453277Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Face synthesis belongs to the cross discipline of pattern recognition and computer vision, it has important scientific significance. This technology meets great application in the fields of security, video conference, human computer interaction and so on. However, natural facial images affected by many kinds of factor have strong diversity in the mode, and the synthesis of multi-pose face images is often a prerequisite and the key step of face recognition, so multi-pose face synthesis is the main task in this paper so that a further work of recognition should be operated in a single image instead of multi-pose images. Because of the significant difference in appearance owned by multi-pose faces, there is great challenge in face synthesis. In this paper, we focus on the difference and propose the following method for multi-pose face synthesis. The author’s major contributions are outlined as follows:1. Active Appearance Model has been studied. This model has been employed to localization of facial feature. Due to its flexible structure and excellent performance, AAM has been extensively applied to the field of computer vision.2. A synthesis method based on LLE algorithm is proposed. With the theory of Active Appearance Model, faces could be divided into shape and texture information to investigate the diversity in poses. Second, Locally Linear Embedding is introduced to purchase the nearest textures to the face which needs to be synthesized and face under different poses could be synthesized with a weighted combination of these neighbors.3. A synthesis method based on pose manifold is proposed. We use pose manifold and linear affine transformation to realize the multi-pose face synthesis. Because of the similarity of pose information in different people, the theory of pose manifold is proper. Finally, multi-pose face synthesis results are given and the effectiveness of the proposed method is verified by evaluation and face fitting which is based on the result in step 2 and the one in step 3.
Keywords/Search Tags:face synthesis, AAM, manifold learning, pose manifold
PDF Full Text Request
Related items