基于Gabor特征和遮挡字典的带遮挡人耳识别  

Ear recognition under partial occlusion based on Gabor feature and occlusion dictionary

作　　者：袁立[1] 刘威[1] 

机构地区：[1]北京科技大学自动化学院,北京100083

出　　处：《仪器仪表学报》2015年第9期2037-2043,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61300075)项目资助

摘　　要：通过引入遮挡字典来编码图像受遮挡部分,稀疏表示分类方法在带有遮挡情况下的人耳识别中能够取得较好的识别性能。然而,常规的利用单位阵作为遮挡字典的方法会对稀疏模型求解带来很大的计算量。提出了一种基于Gabor特征和Gabor遮挡字典的稀疏表示分类方法。利用图像的Gabor特征构造无遮挡字典,因为这种局部特征在姿态变化或遮挡情况下具有一定的鲁棒性。通过学习算法计算出比单位阵遮挡字典更为合理的Gabor遮挡字典,使得图像中被遮挡部分在遮挡字典上的稀疏编码具有更大的稀疏度。在两个人耳图像库上的实验结果表明,相比已有的基于稀疏表示的人耳识别方法,该方法在遮挡情况下能够取得更好的识别效果;对真实环境中存在头发遮挡的人耳识别,也能够取得较好的识别性能。Through introducing an occlusion dictionary to encode the occluded part on the source image, the sparse representation based classification method has shown good performance in ear recognition under partial occlusion. However, the large numbers of atoms in the conventional method using unit array as the occlusion dictionary bring heavy computational burden to the SRC model solving. In this paper, we propose a Gabor feature and Gabor occlusion dictionary based Sparse Representation and Classification （GGSRC） scheme for ear recognition. The non-occlusion dictionary is constructed with the Gabor features, since the Gabor features are more robust to pose variation and partial occlusion. The learning algorithm is used to calculate the Gabor occlusion dictionary that is more reasonable than identity occlusion dictionary and has stronger discrimination power than identity occlusion dictionary, which makes the sparse coding of the occluded part of the image have larger sparsity. The experiment results on two ear image datasets show that compared with the ear recognition method based on SRC, the proposed GGSRC method achieves better recognition results under partial occlusion. In the real case of ear recognition with hair occlusion, the proposed method also shows good performance.

关 键 词：带遮挡人耳识别 稀疏表示 GABOR特征 Gabor遮挡字典 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP392[自动化与计算机技术—计算机科学与技术]