一种使用计算机提取人脸图像特征的阈值最优分层法

在人脸识别中,人脸图像对光照环境变化非常敏感,并严重地影响人脸图像的识别率。针对这一问题,提出了一种使用计算机提取人脸图像特征的阈值最优分层法。该方法依据定义的图象空间分割原理,将图像阈值的选取问题转化为阈值的最优化问题,对所设计的目标函数引入加权系数,并利用单纯形法进行寻优,使问题的求解更符合实际。采用ORL人脸数据库进行仿真实验,利用阈值最优分层法将人脸图象空间划分成不相重叠的子图象空间,使每层子图象空间象素数均匀分布。结果表明,该方法具有阈值自适应调节的特性,避免了固定阈值选取对光照环境变化非常敏感的缺点,对人脸图像的识别具有一定的意义。

A Method for Segmentation of Threshold Value Optimization in Face Image Feature Extraction by Computer

In face recognition,face image is much sensitive to the change of illuminating environments,and recognition rate of face image must be influenced badly.In order to solve this problem,a method for segmentation of threshold value optimization is presented in this paper.According to segmentation principle of image space defined,the method converts the selection problem of image threshold value into the optimization problem of threshold value,introduces weighted coefficients to the objective function designed,and utilizes simplex method in optimization.Thus,the problem solution is more reasonable.Finally,to the face database by ORL,the method is utilized to segment face image space into several non-overlapping subimage spaces so that the pixel number has uniform distribution in each subimage space.Experimental results demonstrate the method has the performance of adaptive threshold adjustment ,and the shortcoming that fixed threshold value selection is sensitive to the change of illuminating environments is avoided.Therefore,it is significant to face image recognition.