融合加速稳健特征的子空间人脸识别方法

为了充分利用人脸特征信息更加准确全面地描述人脸,提高现有识别算法的识别率,提出一种融合改进的加速稳健特征和子空间特征进行人脸识别的方法。利用AAM形状模型的训练方法,训练得到41个点的人脸形状模型;对每幅图像进行特征点初歩定位,找到并保留与初歩定位的特征点空间距离最近的SURF特征点。将SURF特征点描述子利用PCA降维,得到改进的SURF局部特征向量。然后利用PCA对图像进行全局特征提取,将局部特征与全局特征进行融合,组成全新的特征向量。最后通过特征向量的匹配实现识別。对包括本算法和PCA-SIFT算法在内的6种不同识别算法进行了验证。实验结果表明,提出的算法在改变ORL人脸库中训练集样本数的情况下,识别率均优于其他算法;在样本数为5的情况下比PCA-SIFT方法提高了4.3%,可见该算法提高了人脸的识别率具有较强的鲁棒性和分类性。

Subspace face recognition algorithm fused with speed-up robust features

In order to make full use of the information of facial features, describe the human face more accurately and comprehensively, and improve the recognition rate of the existing recognition algorithms, the face recognition algorithm of subspace fused with speed-up robust features was proposed. The face shape model of 41 points was extracted using the training method of AAM shape model, initial position of the face image feature points were obtained by the shape model. The SURF feature points nearest with the initial position of the face feature points were reserved. This paper used PCA to reduce the dimensionality for the reserved SURF feature points, and as the local feature vector of the improved SURF. Afterwards the global feature vector was extracted by PCA, and a new feature vector which was the fusion of local and global feature vector was made up. Finally the face recognition was accomplished by matching feature vector. The improved algorithm and other 5 algorithms which include PCA-SIFT have been tested by ORL database. The experimental results showed that the recognition rate of the improved algorithm was higher than the other algorithms under the changed number of training sample set, and the recognition was improved 4.3% compared with PCA-SIFT under the 5 sample training. So the improved algorithm can effectively enhance the facial recognition rate, and it has a strong robustness and classification.