• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2005, Vol. 27 ›› Issue (9): 51-53.

• 论文 • 上一篇    下一篇

一种基于Snakes方法的人脸五官几何特征提取新算法

肖秀春 邹北骥   

  • 出版日期:2005-09-01 发布日期:2010-07-03

  • Online:2005-09-01 Published:2010-07-03

摘要:

本文对传统Snakes方法的能量函数进行了改进:增加了对Snakes曲线进行形状限定的能量项,用来约束Snakes曲线变化时的形状,使之更容易与相应人脸特征吻合;改进了其中的图像能量项,采用图像颜色色度梯度取代图像灰度梯度构造图像能量项,由于色度梯度体现了颜色的本质信息,这种改进有利于Snakes曲线正确收敛于颜色本质差异大的 的地方即图像真实轮廓处。实验结果表明,改进后的Snakes用于人脸特征提取鲁棒性强。

关键词: 人脸 特征提取 Snakes方法 能量函数

Abstract:

In this paper, we make some improvements on the energy function of the conventional Snakes:(1) we put forward a new energy item (named shape energ y item) to restrain the shape of the Snakes when it moves, this results in better images close to the corresponding facial features; (2) we improve t the image energy item by replacing the gray gradient with the chroma gradient. For the chroma gradient contains substantial information of colour, this improvement is helpful for the Snakes to converge to the correct contour in the image. The results of our experiments show that the improved Snakes is robust in the extraction of facial features.

Key words: (face, feature extraction, Snakes, energy function)