An Unconstrained 3D Expression Transfer Method

Abstract

The invention discloses an unconstrained 3D expression migration method, which is realized by using a method based on computer vision and probability statistics; firstly, adaboost+harr features are used to detect the face area, and then in the face area according to the constrained local model (Constrained Local Model, CLM) method to extract facial geometric features, then use Support Vector Regression (SVR) to extract facial expression parameters, and finally input expression parameters to control the facial BlendShape of the 3D model, synthesize expression animation, and realize an unconstrained 3D expression transfer method.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more particularly, relates to an unconstrained 3D expression migration method. Background technique [0002] Facial expressions play an important role in communication between people. Compared with media such as text and speech, facial expressions have more intuitive and accurate advantages in expressing people's emotions. This emotional interaction mode of people has now been used in scenarios such as virtual reality, digital entertainment, communication and video conferencing, human-computer interaction, etc. Compared with traditional voice, control panel and other interaction methods, it has strong expressiveness and more natural interaction. and other advantages. The expression transfer method roughly includes the following three aspects: facial expression capture, facial expression parameter extraction, and parameterized target facial animation synthesis. [0003] At present, ...

AI Technical Summary

Problems solved by technology

[0004] Most of the expression feature parameterization methods use machine learning algorithms to train the model through a large number of data sets, and learn the mapping relationship between the captured expression information and the expression control parameters, but this type of method is limited by the individual differences of users. The performance results of the method depend to a large extent on the training data and the ability to detect facial expressions in the natural state. To solve this problem, some scholars learn the current user's expression characteristics through additional initialization steps, which effectively weakens the personality differences. , but it increases the complexity of algorithm execution

[0005] Animation synthesis methods can be divided into 2D face animation and 3D model face animation according to the target face type. 2D face animation is based on images, and can obtain high-realistic 2D face animation, but in the synthesized animation, it is difficult to change The lighting conditions of the face and the posture of the face are also difficult to seamlessly stitch into the 3D scene

The muscle model driving method in 3D model facial animation is difficult to obtain control parameters through computer vision algorithms, and the labor consumption of manual control and building the 3D model is very high, and the facial animation method of mixed samples needs to preset the basic expressions. Expression library, the expression library needs to meet the requirements of orthogonality and comprehensiveness, and the artificial cost of building the expression library is very high

Images