Bayesian Tangent Shape Model
To Appear in the IEEE International Conference on Computer Vision and Pattern Recognition, Wisconsin, June 2003
Bayesian Tangent Shape Model:
Estimating Shape and Pose Parameters via Bayesian Inference*
Yi Zhou**
Microsoft Research Asia SMS, Peking Univ.
Abstract
In this paper we study the problem of shape analysis and its application in locating facial feature points on frontal faces. We propose a Bayesian inference solution based on tangent shape approximation called Bayesian Tangent Shape Model (BTSM). Similarity transform coefficients and the shape parameters in BTSM are determined through MAP estimation. Tangent shape vector is treated as the hidden state of the model, and accordingly, an EM based searching algorithm is proposed to implement the MAP procedure. The major results of our algorithm are: 1) tangent shape is updated by a weighted average of two shape vectors, the projection of the observed shape onto tangent space, and the reconstruction of shape parameters. 2) Shape parameters are regularized by multiplying a ratio of the noise variations, which is a continuous function instead of a truncated function. We discussed the advantages conveyed by these results, and demonstrate the accuracy and the stability of the algorithm by extensive experiments.
Lie Gu
Computer Science Department Carnegie Mellon University
Hong-Jiang Zhang Microsoft Research Asia
1. Introduction
The geometrical description of an object can be decomposed into two parts: the geometrical transform and the shape. A common vision task is to recover both pose parameters and low-dimensional representations of the underlying shape from observed images. This procedure is usually referred as “shape analysis” or “shape registration”.
Shape analysis has been advanced in both the literature of statistics and vision. The statistical theory of general shape space began with the work of Kendall [6] in 1977. Kendall described shape distribution in a Riemann
* The work presented in this paper is performed in Microsoft Research Asia and Carnegie Mellon University.
** This author is partially supported by NSFC grant #10171005.
manifold which is highly curved and nonlinear. Statistical techniques were first introduced to analyze the probabilistic distribution of shape in this manifold. Subsequent developments [7][5][2] have led to several practical statistical approaches to analyzing objects using probability distributions of shape and likelihood based inference. A comprehensive survey can be found in Small [5]. General shape space has been proved to be highly nonlinear. However, as for a set of concentrated data, tangent space provides a good linear approximation to general shape space. More importantly, modeling shape in tangent space can convert statistical shape analysis to standard multivariate analysis [7].
In image analysis literatures, practical parametric deformable models [3][10][11] have been developed to deal with the problems like segmentation or feature points localization. These models are generally capable of incorporating prior knowledge with observations directly derived from image data. In particular, Active Shape Model [3] proposed by Cootes et.al. in 1992 attracts a wide range of attention. ASM consists of a point distribution model capturing shape variations of valid object instances, and a set of grey gradient distribution models, which describe local texture of each landmark point. Cootes developed an iterative searching algorithm to actively update the model parameters according to the observed image. The major advantage of ASM is that the model can only deform in the ways leant from the training set. That is, it can accommodate considerable variability and it is still specific to the class of object it intends to represent. Specifically, in ASM the principle component analysis (PCA) technique is used to model both 2D shape variations and local grey level structures.
In this paper, we address the problem of shape analysis from two aspects. First, shape analysis problem is formulated in Bayesian framework. Specifically, we describe the prior model of tangent shape vectors, the likelihood model and the posterior of model parameters. Second, an EM based searching algorithm is given to estimate tangent shape and other model parameters. The
Bayesian Tangent Shape Model For Face Alignment 准备知识 ? ASM算法 ? EM算法 ASM(1/7) ? 一个物体的几何描述分为两部分:– 相似变换(旋转、缩放、平移) ...
2003年Leon Gu[4]等在此基础 上提出了BTSM(Bayesian Tangent Shape Model)进行人脸轮廓提取,其基本思想是 通过对ASM中的残差部分建立模型, 可以得 到更加准确的...
Zhou, L. Gu, H.J. Zhang. Bayesian Tangent Shape Model - Estimating Shape and Pose Parameters via Bayesian Inference [C]. In Proc. of IEEE Conf. ...
Leon Gu, Bayesian Tangen... 8页 1财富值 RELATIONS BETWEEN LIN... 暂无... A detailed description of the development of the tangent linear model (TLM...
25. Zhou Y., Gu L., Zhang H-J, Bayesian tangent shape model: estimating shape and pose parameters via Bayesian inference, ISBN: 0-7695-1900-8, ...
shape model (TC-ASM) to localize a face in ...tangent space of the mean shape S: After s ...3.2. TC-ASM in Bayesian framework TC-ASM ...
shape model (TC-ASM) to localize a face in ...tangent space of the mean shape S: After s ...3.2. TC-ASM in Bayesian framework TC-ASM ...
热门文档