摘要
为了提高ASM在非均匀光照条件下的人脸面部特征点定位的精确度,提出了一种融合Log-Gabor小波特征的多模型ASM方法.该方法的主要特点有:在精确定位目标图像虹膜位置的基础上对全局形状模型进行较准确的初始化;特征点局部纹理特征采用灰度和Log-Gabor小波特征共同描述,减少光照和噪音对算法的影响;建立包括全局ASM和基于人脸面部显著特征区域的局部ASM的多模型ASM,交替使用这两种ASM模型在边缘约束策略基础上对特征点的定位结果进行约束.实验表明,多模型ASM算法对人脸面部特征点定位的准确率比传统ASM算法有明显提高.
Active shape model is one of the most popular methods for facial feature point location. To improve its accuracy with variance expressions and under non-linear illumination, a multi-model ASM method which integrates Log-Gabor wavelet features is proposed. Irises were located accurately firstly and then utilized to initialize the global ASM; the gray and Log-Gabor wavelet were used to depict the character of each feature point to decrease the effect of illumination and noise, global ASM and local ASM were built respectively and used to constrain feature point location alternately. Experimental result shows that multi-ASM can achieve much higher accuracy than the traditional ASM algorithm.
出处
《微电子学与计算机》
CSCD
北大核心
2008年第6期105-107,111,共4页
Microelectronics & Computer
基金
国家自然科学基金项目(60475021)
