基于模板匹配和遗传算法的人眼定位

An Eye Orientation Approach Based on Deformable Template and Genetic Algorithms

文中提出了一种利用模板匹配与遗传算法相结合的人眼定位算法。根据人脸几何特征将人脸分为几个特征区域(眼睛、鼻子、嘴巴),在找到人眼区域后根据人眼几何特征,建立人眼特征模板及其能量函数,通过改变模型参数用遗传算法实现能量函数的全局最小化,使能量函数最小的模型参数即可描述人眼的特征,从而实现人脸特征的有效定位。同时还讨论了不同的能量函数在人眼定位过程中的地位和作用。可避免在能量函数寻优过程中的局部最小化的出现,较准确较快捷的确定人脸主要特征的位置。

A method of extracting eye features from facial images using deformable templates and genetic algorithms is described in this article. Using this method the orientation of eye features can be completed. According to facial geometrical structure the facial images are divided into several areas and each area includes one feature of facial images. After finding the eye areas the eye features template and its energy function are constructed. By altering its parameter values to minimize the energy function, the best fit can describe the eye features. In this process the genetic algorithms are used to realize minimum of energy function. At the same time the effects of different energy functions on eye orientation are also discussed. This method can avoid the appearance of local minimum in energy function matching course and effectively complete the orientation of facial features .