结合Viola-Jones人脸检测框架和CLM(Constrained Local Model)模型,建立了一种对光照和姿态变化鲁棒的人脸检测算法.该算法首先对不同状态下的训练集使用AdaBoost算法建立一组状态相关的强分类器和相应的CLM模型.然后通过采用多层级联结构分类器对图像中存在的人脸进行检测并对检测到的人脸进行光照和姿态变化的分析.最后,选择合适的CLM模型对检测到的人脸进一步进行定位,从而得到精确的人脸位置信息.通过在CMU-PIE人脸数据库上进行实验,验证了该算法的有效性.%A robust and accurate face detection algorithm to illumination and pose variations based on Viola-Jones face detection framework and Constrained Local Models (CLM) is proposed. This algorithm builds a set of variation-specific strong classifiers using AdaBoost algorithm and corresponding CLM for different illumination and pose variations. Then a multi-layer cascade classifier is adopted to determine the rough location of the face region and estimate the illumination and pose conditions of each face in input images. At last, a suitable CLM is selected to obtain accurate landmarks of faces in input images. Experimental results on CMU PIE face database show the efficiency of this proposed algorithm.
展开▼
