改进的用于人脸面部特征定位的ASM方法

主动形状建模是面部特征定位和人脸识别等模式识别领域中常用的一种方法.然而,由于受到初始情况、光照等诸多因素的影响,主动形状建模经常会陷入最优化过程中的局部最小问题,从而导致其性能下降.文章在传统主动形状模型基础上,提出了一种改进的ASM算法.首先,利用Adaboost初始定位面部的显著特征区域以便在后续的搜索中进行位置的约束;其次,将原始ASM方法中的关键点的1D纹理模型改进为基于核概率密度估计模型的2D纹理模型;最后,在局部灰度模型中加入了边缘约束,使边缘信息较强的点有更大的可能成为最佳候选点.该方法可以有效地解决上述局部最小问题,并且更好地捕捉局部点的特征信息,从而更精确地进行面部特征定位.实验结果表明,改进的ASM方法在准确性和鲁棒性上有较大提高,可为后续的人脸识别打下良好的配准基础.

Improved ASM for Facial Feature Location

Active Shape Models(ASM) is one of powerful tools for facial feature location and face recognition.However,the performance of ASM is often influenced by some factors such as the initial location,illumination and so on,which will frequently lead to the local minima in optimization.This paper proposes an improved ASM algorithm on base of traditional one.First,using Adaboost method initial positioning of the significant facial region in order to carry out follow-up search in the location of the binding.Second,...