针对视频的人脸卡通化方法研究

发布时间：2017-12-31 10:27

本文关键词：针对视频的人脸卡通化方法研究　出处：《电子科技大学》2016年硕士论文　论文类型：学位论文

【摘要】：近几年来,动漫产业呈现爆炸式的增长,卡通人物形象更是被创造性地应用在视频通话、角色游戏等方面,这些都离不开计算机辅助动画合成技术在降低工作量、提高表现效果方面发挥着的关键性作用。其中,视频人脸卡通化一方面要考虑视频的实时性要求;另一方面,人脸表情变化丰富细微,要兼顾生成卡通与真实人脸的相似性,因此人脸卡通化一直是计算机辅助动画合成技术面临的重点和难点。现有的视频卡通化方法需要复杂的预处理或大量的后期工作,难以实时地生成个性化的卡通人脸。本文围绕视频人脸卡通化这个问题进行了深入研究,在此基础上,进一步研究了基于平均脸的人脸夸张化方法,具体研究内容如下:1.提出了改进的交互式人脸卡通化方法,该方法能够实现在复杂背景下合成效果较好的卡通人脸,仅需要简单的用户交互。本文首先提出了基于回归树特征点模型的人脸卡通合成方法,针对人脸卡通轮廓线合成效果不好的问题,提出基于二次B样条插值的修正方法;针对复杂背景下头发区域分割效果不好的问题,提出了分块闭操作方法。2.提出了一种人脸视频卡通生成方法,包括基于器官状态的相似帧查找以及基于回归树特征点模型的卡通驱动。针对基于灰度值的相似帧查找错误率高的问题,提出以器官的状态作为两帧相似的判定标准,改善了查找的准确性。基于回归树模型的卡通驱动将视频中器官的运动分解为刚性运动和柔性运动,通过对视频的压缩处理提高特征点的检测速度,并根据检测到的器官运动参数驱动相应的卡通器官完成对两种运动的响应。3.提出了基于平均脸的两种人脸夸张化方法:人脸整体夸张方法和针对突出器官的局部夸张方法。前者分别构造并比较测试人脸与平均人脸之间的特征向量,对所有存在差异的部分使用薄样条插值函数进行夸张变形;后者根据欧氏距离得到最突出的器官,并对测试人脸的脸型进行判定,最后仅对最突出器官及脸型进行夸张变形。4.视频人脸卡通化系统开发。该系统能够通过简单的人机交互实现视频人脸卡通化算法以及对人脸的夸张化算法。
[Abstract]:In recent years, the animation industry shows an explosive growth, cartoon characters are creatively used in video calls, character games and so on. These are all inseparable from the computer aided animation synthesis technology in reducing the workload and improving the performance of the key role played by the video face card on the one hand to consider the real-time requirements of video; On the other hand, the facial expression is rich and subtle, so the similarity between cartoon generation and real face should be taken into account. Therefore, face card generalization has always been the focus and difficulty of computer-aided animation synthesis technology. The existing video card generalization methods need complex preprocessing or a lot of later work. It is difficult to generate personalized cartoon face in real time. This paper focuses on the problem of video face card generalization. On this basis, the method of facial exaggeration based on average face is further studied. The specific research contents are as follows: 1. An improved interactive face card generalization method is proposed, which can synthesize cartoon face with better effect in complex background. Only simple user interaction is required. Firstly, a face cartoon synthesis method based on regression tree feature point model is proposed. A modified method based on quadratic B-spline interpolation is proposed. In order to solve the problem that the segmentation effect of hair region in complex background is not good, a block closed operation method is proposed. 2. A face video cartoon generation method is proposed. It includes similar frame lookup based on organ state and cartoon driver based on regression tree feature point model. Aiming at the problem of high error rate of similar frame lookup based on gray value. Using the state of organs as the criterion of similarity between two frames, the accuracy of searching is improved. The motion of organs in video is decomposed into rigid motion and flexible motion by cartoon driver based on regression tree model. Video compression improves the detection speed of feature points. According to the detected organ motion parameters, the corresponding cartoon organs are driven to complete the response to the two movements. 3. Two face exaggeration methods based on average face are proposed. The whole exaggeration method of face is constructed and compared with the local exaggeration method for salient organs. The feature vectors between the tested face and the average face are constructed and compared respectively. For all the different parts, the thin spline interpolation function is used for hyperbolic deformation. The latter gets the most prominent organs according to the Euclidean distance and determines the face shape of the tested face. Finally, only the most prominent organs and facial patterns are exaggerated. 4. The video face card system is developed. The system can realize video face card generalization algorithm and face exaggeration algorithm through simple human-computer interaction.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】