Convergence Deformation Model And The Irradiation Constraint Mechanism Of Facial 3 D Shape Recovery

With the rapid development of computer visual applications,3D facial reconstruction has become a hot research direction and achieved certain results because of its broad application prospects,. At present, there are two methods which attract the attention of the vast number of researchers:Morphable model method and Shape-from-Shading (SFS) method. Morphable model method is modeling based on three-dimension facial samples, and then modify model according to specific two-dimension facial image; SFS method uses two-dimension facial image to abstract image gray-scale information directly and then recovery three-dimension facial shape.Because of the difference of the above two method’s principal, they have different advantages and disadvantages.Though the accuracy of three-dimension face using morphable model method is very high, the detail of facial characteristics parts (such as:eyes, dimples, scars, etc) are poorly recovery. The most important is that this method is not suitable to real-time applications due to its high complexity of computational time.On the other hand, though arithmetic operation of SFS is very quick, the shape of recovery using SFS is to be improved, because lots of shape information will be lost in the process of image generation.Both methods have their own advantages, so this article proposes the algorithm, which combines morphable model method and SFS in the constraint of image irradiance in order to improve the accuracy of the algorithm. We get initial facial shape by making full use of morphing modeling, and then use SFS to facial normal vector to enhance the precision of recovery3D face. In this way, we can recovery normal vector which contains high-resolution facial detail features (such as wrinkles, dimples, scars, etc.), and these details cannot be obtained by directly using morphing modeling. The detail obtained in this way will be very helpful to improve the accuracy of the facial shape in the new direction of the lighting source.In addition, we will study a new method, which is used to estimate the albedo according to the different areas, rather than using unified global albedo, and at the same, use a hybrid reflection model in order to enhance the accuracy of the recovery facial shape and improve adaptability of algorithm under different lighting conditions.