| 英文摘要 | With the development of digital media and animation industry, in some cases, people are not satisfied with using real image, but pursue cartoon of the real one. How to turn real facial photos to cartoon ones is receiving common attentions. In the real facial photos, the person’s expression can be one of the prototypic emotional expressions, i.e., neutral, happy, grief, disgust, surprise, angry and fear, however, such prototypic expressions occur relatively infrequently, sub-expressions are occurred more frequently. In order to grasp the emotion state of real face better, Paul Ekman et al develop the Facial Action Coding System (FACS), which describes facial expressions by 41 action units. These 41 AUs and AU combinations can express a variety of complex expressions. In this paper, based on analysis and recognition of the input real facial image, we propose a facial feature and action units-based cartoon system, which can generate a realistic cartoon face form an input image; The system can also recognize action units of the input one, and generate a cartoon face that owns the specific action units. Specifically, in this paper, the work that has been finished includes following four aspects: (1) Analyze facial images of different quality, find the robust facial feature points locating technology, and improve the accuracy and applicability of locating technology in model training and algorithm performance. (2) Analyze action units that prototypic expressions and sub-expressions display, access the frequently occurring action units, and find pattern classification method that can recognize above action units robustly. (3) Utilize Delaunay triangulation in computer graphics to do cartoon texture mapping, and make use of image mapping and curve drawing to add ears and hair to cartoon face. The system can generate different cartoon texture and hair by different cartoon texture template and hair-style template. (4) For a non-neutral expression face image, the system can identify the action units of its performance，and generate cartoon face that owns the specific action units by cartoon generation algorithm. |
修改评论