摘要
本文提出了一种基于复合局部信息模型的改进Active Shape Model(ASM)算法,并进一步提出了基于人脸特征点Gabor小波特征降维分类的特征点搜索方法,对改进ASM的结果进行精确校正,达到鲁棒精确定位特征点的目的.本文首先对经过Adaboost检测的彩色图像人脸区域进行光照补偿,然后根据眼睛和唇色的色度特性定位双眼和嘴唇中心位置,从而得到较好的人脸特征点的初始位置.在特征点位置搜索中,本文结合肤色概率信息对ASM方法进行了改进,从而提高了仅基于灰度梯度信息的传统ASM方法的鲁棒性和准确性.最后选取改进ASM搜索后的某些特征点一定领域内梯度值较高的点,提取其Gabor小波特征,通过线性判别式分析法(Linear Discriminant Analysis)和最近邻分类法对其进行分类,搜寻最符合训练样本Gabor特征的点作为最佳位置点,修正了ASM的搜索结果,使得搜寻结果更加精确.
We present a method of facial feature point extraction based on improved Active Shape Model （ASM） and Gabor wavelet. Facial feature points can be located robustly and precisely by using the method proposed in this paper. Firstly, fight compensation is used on the detected facial region. Secondly, we locate the eyes and mouth to obtain a good initial shape of feature points. And then the improved ASM is used to get a robust tough shape. We improve local gray-level model of the traditional ASM by adding local skin similarity probability,and form local hybrid model. Finally, we search the precise location of a certain feature point by classifying the Gabor feature of points around the rough location. The Gabor feature can be trained and dimensionally reduced by Linear Discriminant Analysis （LDA） ,and classified by Nearest Neighborhood method. At the end of the paper, experimental results demonstrate the robustness and accuracy of this method.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2008年第2期309-313,共5页
Acta Electronica Sinica
基金
国家自然科学基金(No.60772097)
关键词
人脸特征点定位
肤色概率模型
GABOR特征
特征点分类
线性判别式分析法
最近邻分类
facial feature points extraction
skin similarity model
Gabor feature
feature point classification
linear discriminant analysis LDA： nearest neighbor classification
