基于Gabor滤波器的快速人脸识别算法

摘要:针对传统人脸识别方法中所提取特征维数高、计算量大等缺点，提出一种新的正面人脸识别算法。新算法融合了半边人脸识别方法、Gabor滤波器、基于互信息判据的Gabor特征筛选来进行人脸识别。新算法将人脸图像分为左右两个部分，计算并比较人脸图像左右半边脸的熵，选取熵值较大的半边人脸图像进行Gabor特征提取。利用二值分类器判别单个Gabor特征的分类能力，选取分类能力较强的特征（最具判决力的特征）。再利用互信息判据对Gabor特征进行第二次筛选，以减小特征之间的冗余度。最后利用最近邻判别器来进行人脸识别。实验结果表明，新算法的识别率优于传统半边脸识别方法，识别速度也优于传统的利用Gabor滤波器进行特征提取的方法。

关键词:人脸识别；Gabor滤波器；特征选择；互信息

中图分类号: TP391.41文献标志码:A

New fast face recognition algorithm based on Gabor filter

KONG Rui*, HAN Ji.xuan

College of Electrical and Information, Jinan University, Zhuhai Guangdong 519070, China

Abstract:

Aiming at disadvantage of classical Face Recognition Algorithm, such as extracted feature dimension is higher, compute cost is huge. A fast face recognition algorithm is presented. The algorithm comes from the integration of the half face recognition scheme, the Gabor filters, Gabor features selected method based on mutual information, and the nearest neighbor method for frontal face recognition. The face images in training set and testing set are divided into the left half and the right half, one half of the face images is chosen by entropy maximum. The features of the face images are extracted by Gabor filters. Then the rank of discriminating capabilities of features can be estimated by evaluating the classification error on intra-set and extra-set based on weak classifier built by single feature．The Gabor features with small errors were selected．And at the same time, the mutual information between the selected features was examined．The nearest neighbor method is used to recognize the frontal face. The experimental results show that the proposed method has higher accuracy than the traditional half face recognition algorithm, has lower computational complexity than the traditional Gabor filter algorithm.

Concerning the disadvantage of traditional face recognition algorithm, such as high dimension of extracted feature, a great deal of computation, a fast face recognition algorithm was proposed. The algorithm integrated the half face recognition scheme, Gabor filter, Gabor features selecting method based on mutual information, and the nearest neighbor method for frontal face recognition. The face images in training set and testing set were divided into the left half and the right half, one half of the face images was chosen by entropy maximum. The features of the face images were extracted by Gabor filter. Then the rank of discriminating capabilities of features can be estimated by evaluating the classification error on intra.set and extra.set based on weak classifier built by single feature．The Gabor features with small errors were selected．And at the same time, the mutual information between the selected features was examined．The nearest neighbor method was used to recognize the frontal face. The experimental results show that the proposed method has higher accuracy than the traditional half face recognition algorithm, and is of lower computational complexity than the traditional Gabor filter algorithm.

Key words:

face recognition; Gabor filter; feature selection; mutual information

0 引言

人脸识别通过计算机分析人脸图像，并从人脸图像中提取有效信息进行身份识别。它是模式识别领域的重要课题，在门禁、安防等方面有着广泛的应用。在进行人脸识别时，关键的步骤是特征提取，对提取的特征主要有以下两方面要求：1）不同人脸的特征具有较大的区分度，同一人脸的特征具有较大的关联度，即所提取的特征能保证类内散度小，类间散度大；2）尽可能地减小所提取的特征维数，便于快速识别。Gabor于1946年将短时傅里叶变换的窗函数取为高斯函数，提出了Gabor变换。二维Gabor滤波器（函数）是由Daugman首次提出［1］，可以看成是一个高斯函数调制的复正弦函数，二维Gabor函数是唯一能够达到测不准原理下界的函数（测不准原理：不可能在时域和频域都能获得任意的测量精度，要使频率分辨率提高，必然牺牲时域分辨率），即Gabor函数可以同时获得较高的时域和频域分辨率。二维Gabor滤波器与哺乳动物视觉皮层简单细胞二维感受野剖面非常相似，具有优良的空间局部性和方向选择性，能够抓住图像局部区域内多个方向的空间频率和局部性结构特征［2］，对光照与表情的变化具有良好的鲁棒性，利用Gabor滤波器所提取的特征进行识别时，识别率较高。但Gabor滤波器所提取的特征维数过大，且特征提取时运算量过大，导致识别速度较慢。

目前利用Gabor滤波器进行特征提取的主要研究工作是在保证识别率的前提下，尽量减少运算量，压缩特征维数，提高识别速度［3-4］。为了有效地压缩特征维数，文献［5-6］都提出了一种基于互信息量判据的特征压缩方法，首先将特征集中任意两个Gabor特征相减得到“类内差空间”与“类间差空间”，这就将多类问题转化为两类问题，然后研究单个Gabor特征的分类性能，再用特征间的相互信息量，优选出具有无冗余、低误差率的特征子集。该方法在保证识别率与鲁棒性的情况下有效地降低了Gabor特征的维数，但其降维后的Gabor特征仍然有200维左右，此外，该方法需要用Gabor滤波器与人脸库中所有图像卷积，并不能有效地减小运算量，故识别速度并无明显提高。文献［7］提出了一种半边人脸识别的方法，把训练集的人脸图片分为左右两个部分，经过亮度补偿后分别利用主成分分析（ Principal Component Analysis，PCA）特征提取方法提取训练集中左右半边脸的特征，计算待识别图片左右半边脸的熵，选取熵较大的半边脸及对应的训练集特征，经过亮度补偿后提取该半边脸的特征，并根据提取的特征进行分类。这样虽有效地解决了左右脸光照不均的问题，以及对整幅图像进行特征提取运算量过大的问题，但由于其使用的特征提取方法和亮度补偿函数过于简单，导致其对局部亮度不均的图像识别效果不佳。

针对上述问题，本文提出了一种基于Gabor滤波器的快速人脸识别算法，在特征提取方面融合了半边人脸识别方法与Gabor滤波器，把测试人脸图像和训练集人脸图像分为左右两个部分，计算并比较待识别图像左右半边脸的熵，选取熵值较大的半边脸及对应的训练集中的半边人脸图像进行Gabor滤波器特征提取；在特征降维方面，利用二值分类器判别单个Gabor特征的分类能力，选取分类能力较高的特征；再利用互信息判据对Gabor特征进行第二次筛选，以减小特征之间的冗余度；最后利用最近邻判别器来进行人脸识别。