Object Modeling and Analysis Based on Morphable Model: A Review
Hui ZHANG,Guang-you XU,Feng XIE
DOI: https://doi.org/10.3321/j.issn:0254-4164.2001.06.001
2001-01-01
Jisuanji Xuebao/Chinese Journal of Computers
Abstract:Object modeling has been an active research topic since it has wide applications such as recognition, virtual reality, computer animation, etc. In this field, the traditional representation is a 3D wire-frame model, or 3D continuous surface, mapped with some texture. The shortcoming of this kind of model is its computational complexity in model creation and display, data redundancy, and difficulty in realistic rendering. These years morphable model has received considerable attention in modeling an object class(especially human faces) for its efficient data compression and realistic rendering. Its occurrence is inspired by a traditional rendering method in computer graphics, morphing, which generates new images as the interpolation of several given frames according to the correspondences between them. Similarly, morphable model is introduced to represent an object class with the linear combination of some typical instances of this class. This new model represents a new instance as a coefficient vector of the combination, so it is more efficient. Furthermore, such a vector serves as a good feature for object recognition. With this model, object analysis can be converted to a common optimization problem, then more kinds of input data and mathematical tools are ready for utilization. At the same time, image rendering is much simplified. There are two key problems for working with this model. One is to find correspondences between different instances; the other is applying proper constraints during combination in order to avoid generating false instance. Many kinds of application, such as very low bit-rate coding, personal identification, reconstruction, benefit from this model. The author reviews the morphable model and its development and applications, and presents the possible future research orientations.