模型基编码中的鼻子形状自动提取  

Automatic Extraction of Nose Shape in Model-Based Coding

作　　者：杨承根[1] 姚庆栋[1] 

机构地区：[1]浙江大学信息与通信工程研究所,杭州310027

出　　处：《上海交通大学学报》2007年第8期1324-1329,共6页Journal of Shanghai Jiaotong University

摘　　要：介绍了一种简单而有效的模型基编码鼻子轮廓提取算法.该算法利用皮肤颜色图像分割、特征保留图像二值化、特征脸校验、脸部特征滤波和k-均值分类器定位复杂场景中的人脸及脸部各特征区域,提出了以圆形和非对称双抛物线建立的鼻子形状形变模板以及相应的能量函数,寻找的Powell能量最小化算法的初始参数加速了模板匹配过程,给模型基编码的人脸线框模型自动调整提供了准确的鼻子形状参数.实验表明,所提出的算法对不同人脸姿势的鼻子形状估计表现出良好的结果.A simple but effective and applied algorithm for nose shape extraction in model-based image cod- ing was presented. It locates the face region as well as facial feature regions within complex scene by skincolor image segmentation, binary-conversion of image with feature preserving, eigenface verification, facial feature filtering and k-means classifier. It proposes the deformable templates of nose shape described as two circles and two unsymmetrical parabolas and the corresponding energy function, and the detected initial parameters of the nose shape accelerate the template matching by energy minimizing of Powell algorithm. The extracted nose shape provides important feature parameters for the automatic adaptation of general facial wire-frame model in model-based coding. The experiments show that the proposed algorithm performs well in estimation of the nose shape under different face poses.

关 键 词：模型基图像编码 形变模板 特征脸 POWELL算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]