基于K-means的精确人脸对齐算法

Accurate face alignment algorithm based on K-means

针对传统人脸对齐算法效率较低,在人脸表情、头部姿势、光照差异较大的情况下鲁棒性差等问题,提出一种基于K-means的精确人脸对齐算法。首先,针对训练图像,采用K-means实现聚类,将训练图像分为k类,使距离相近的人脸图像聚为一类。其次,针对输入图像,找到与其相似的类,通过K最近邻(K-NN)算法,选取k张与输入人脸图像最相似的训练图像,建立输入图像的形状和外观模型。然后,将非线性脸部模型转换为一系列的线性组合实现快速拟合。最后,通过300—W基准数据集测试,测试结果表明,与SDM和ESR人脸对齐方法相比,K-means的人脸对齐算法精度提高了2%~4%。

An accurate face alignment algorithm based on K-means is proposed,aiming at the problems of low efficiency of traditional face alignment algorithm and poor robustness in the case of facial expression,head posture and illumination.Firstly,for the training images,the K-means is used to realize clustering.Training image is fell into K,and close by face image is classified one category.Then,the most similar category as the bases of shape and appearance model through K-nearest neighbour(K-NN)algorithm for the input image.Nonlinear face model is transformed into a set of linear combinations to achieve fast fitting.Finally,to validate the effectiveness of our algorithm,comprehensive experiments are conducted on 300-W dataset test.Test results demonstrate that the proposed algorithm is improved by 2%~4%compared with SDM and ESR methods in terms of precision.