摘要
针对人脸识别的预处理,采用图像处理技术解决了人脸检测问题。首先建立输入图像的肤色模型,然后进行开运算处理,以消除图像噪声利于后面的眼睛定位。再对二值图像做灰度投影实现人脸粗分割,定位双眼。最后对细化分割出来的人脸区域进行标准化操作,包括灰度的均衡处理和Mallat算法二维小波分解。灰度均衡把原始图像的直方图变换为均匀分布的形式,增加像素灰度值的范围。小波分解可以压缩图像,以降低算法的复杂度。每个步骤通过处理前后人脸图像的对比彰显所做步骤的意义。人脸检测的最终结果是获得64×64大小的人脸图像。此图像包含了人脸的有效信息,在此图像的基础上才能进行后续的提取特征、设计支持向量机,进而做人脸识别。
According to the pretreatment of face recognition, by using image processing technology the problem of face detection was solved.At first the input image model of skin color is established, and opening operation is used to eliminate image noise for the eyes locating.The binary image is projected in gray-scale to realize rough face segmentation and eyes location.The standardized operation is carried out on the face region of subtle seg-mentation, including gray-scale equalization and 2D wavelet decomposition by Mallat algorithm.It increases the range of the pixel gray-scale that gray-scale equalization transforms the histogram of original image into uni-form distribution’ s shape.It reduces the algorithm complexity that wavelet decomposition compresses image .Ev-ery step it shows the images contrast between before and after processing to reveal the purpose of every step.The final result of face detection gets a face image of 64 ＆#215;64.This image includes effective information of the face. Based on this image the feature extraction, support vector machine design and then face recognition can be car-ried out.
出处
《安徽理工大学学报（自然科学版）》
CAS
2014年第3期33-38,共6页
Journal of Anhui University of Science and Technology:Natural Science
基金
安徽省高等学校省级优秀青年人才基金重点资助项目(2013SQRL104ZD)
关键词
图像处理
人脸检测
灰度均衡
小波分解
肤色建模
image processing
face detection
gray-scale equalization
wavelet decomposition
skin color modeling
