改进的SIFT特征人脸识别方法ADSIFT

An Improved Method Based on SIFT Feature for Face Recognition

针对光照、表情、噪声等因素容易造成误识别的问题,提出一种改进的SIFT特征人脸识别方法.对每个训练图像,先提取得到SIFT特征向量集合,利用每个SIFT特征向量,并选择阈值构造一个弱分类器.利用一种基于Adaboost的算法从每个训练图像的弱分类器集合中选出一部分,确定其对应的阈值和权重,然后构造出该训练图像的相似度函数.根据相似度函数可计算出目标图像与每个训练图像的相似度,从而求出目标图像与每个类的训练图像的平均相似度,则目标图像属于平均相似度最高的类.实验表明在ORL人脸数据库上则可达到98%识别率,优于现有的方法.

By aiming at the problem of misrecognition caused by factors such as light,expression and noise,an improved method based on SIFT feature for face recognition is proposed.First of all,a set of SIFT feature vectors are extracted from every training image.A weak classifier can be constructed with a SIFT feature vector and a threshold.For every training image,some weak classifiers are selected with thresholds and weights by an algorithm based on Adaboost.A similarity function for training images is established.Similarity between object image and training images can be calculated by similarity functions.The average similarity between object image and each class is gained.Finally,an object image is classified to the class with the maximum average similarity.It’s verified that this method is able to increase the recognition rate to 96.82% on AR face database and to 98% on ORL face database,which is better than other existing methods.