摘要
设计了一种复杂背景中的人脸检测与验证方法。在预处理部分该方法使用了线性光照拟合的预处理方法以减轻光照的影响。在检测过程中该方法引入了一种 3× 3的划分方式 ,该划分方式下根据人脸器官的灰度分布特性设计了相应的检测规则 ,并结合改进后的 4× 4划分方式下的人脸检测规则构成了最终的人脸检测规则。经过人脸检测过程后 ,对所得的结果使用具有良好抗噪声性能的最小同值分割吸收核区 (Smallest univalue seg-mentassimilating nucleus,SUSAN)方法进行检测结果验证 ,进一步增强了系统的整体性能。最后通过复杂背景下的人脸图片以及叠加噪声后的人脸图片的检测结果说明该方法具有较高的检测率及良好的抗噪声性能。
A method used to detect and verify the face regions in complex background is proposed. In pretreatment, the technique of linear lightness fitting is used to reduce the influence of the lighting variance. In detection proceeding, a 3×3 face region partite technique is introduced and correspondent detection rules are designed. Combining these rules with the face detection rules corresponding to the 4×4 face region partite technique, the final face detection rules are obtained. Using these rules, the method can detect faces on the image. After the faces are founded, the smallest univalue segment assimilating nucleus method (SUSAN), which has a good attribute of noise resistance,is used to verify the results. It also improves the performance of the system. Finally, the method is used to detect faces in complex background or contaminated noise images and to get good results. Results demonstrate that it has a good detection rate and characteristics of the noise resistance.
出处
《数据采集与处理》
CSCD
2004年第1期10-15,共6页
Journal of Data Acquisition and Processing
关键词
人脸检测
模式识别
计算机视觉
区域特征
图像识别
face detection
smallest univalue segment assimilating nucleus
linear lightness fitting
