Face recognition systems have been in the active research in the area of image processing for quite a long time. Evaluating the face recognition system was carried out with various types of algorithms used for extract...Face recognition systems have been in the active research in the area of image processing for quite a long time. Evaluating the face recognition system was carried out with various types of algorithms used for extracting the features, their classification and matching. Similarity measure or distance measure is also an important factor in assessing the quality of a face recognition system. There are various distance measures in literature which are widely used in this area. In this work, a new class of similarity measure based on the Lp metric between fuzzy sets is proposed which gives better results when compared to the existing distance measures in the area with Linear Discriminant Analysis (LDA). The result points to a positive direction that with the existing feature extraction methods itself the results can be improved if the similarity measure in the matching part is efficient.展开更多
This paper presents a method which utilizes color, local symmetry and geometry information of human face based on various models. The algorithm first detects most likely face regions or ROIs (Region-Of-Interest) from ...This paper presents a method which utilizes color, local symmetry and geometry information of human face based on various models. The algorithm first detects most likely face regions or ROIs (Region-Of-Interest) from the image using face color model and face outline model, produces a face color similarity map. Then it performs local symmetry detection within these ROIs to obtain a local symmetry similarity map. The two maps and local similarity map are fused to obtain potential facial feature points. Finally similarity matching is performed to identify faces between the fusion map and face geometry model under affine transformation. The output results are the detected faces with confidence values. The experimental results demonstrate its validity and robustness to identify faces under certain variations.展开更多
为解决目前深度仿造检测方法对于跨数据集的检测性能难以提高的问题,提出基于注意力机制和一致性损失相结合的深度伪造人脸检测方法(method based on attention mechanism and consistency loss,MAMCL)。采用多注意力机制,迫使网络捕捉...为解决目前深度仿造检测方法对于跨数据集的检测性能难以提高的问题,提出基于注意力机制和一致性损失相结合的深度伪造人脸检测方法(method based on attention mechanism and consistency loss,MAMCL)。采用多注意力机制,迫使网络捕捉到更细微的局部异常。采用基于注意力机制的擦除方式,鼓励模型深入挖掘之前忽略的区域。设计一致性模块获取伪造图像中普遍存在的不一致细节特征,并应用一致性损失引导模型更加关注伪造细节。在面部取证++(FaceForensics++,FF++)数据集上进行实验,准确率达到96.38%,受试者工作特征曲线(receiver operating characteristic curve,ROC)的曲线下面积达到99.34%,在泛化性能测试中也取得了良好的效果。通过消融实验,证明了每个模块的有效性。结果表明,提出的检测方法能够较为准确地检测深度伪造人脸,且具有良好的泛化性能,可以作为应对当前人脸伪造威胁的有效检测手段。展开更多
近年来,稀疏表示分类(Sparse Representation Based Classification,SRC)方法在人脸识别中受到越来越多的关注。原始SRC方法使用所有的训练样本组成字典矩阵,当训练样本比较多时,稀疏系数的求解会变得非常耗时。为了解决这一问题,提出...近年来,稀疏表示分类(Sparse Representation Based Classification,SRC)方法在人脸识别中受到越来越多的关注。原始SRC方法使用所有的训练样本组成字典矩阵,当训练样本比较多时,稀疏系数的求解会变得非常耗时。为了解决这一问题,提出一种新的局部稀疏表示分类(Local SRC,LSRC)方法。该方法针对每个测试样本,根据测试样本和训练样本稀疏系数之间的相似性来选择部分训练样本,由这些训练样本组成字典,然后在这个字典上对测试样本进行稀疏分解。该方法性能相比于原始LSRC方法更稳定。在ORL、Yale和AR人脸库上的实验结果表明,该方法的效果优于SRC和LSRC。展开更多
主要研究了移动平台上的相似脸检索问题.对于移动端,首先采用基于稀疏约束的级联回归模型进行精确的人脸配准,该方法不但能够筛选鲁棒的特征,而且可以将模型的大小压缩到原来的5%左右.接着,在某些关键点周围提取高维的纹理特征,并通过...主要研究了移动平台上的相似脸检索问题.对于移动端,首先采用基于稀疏约束的级联回归模型进行精确的人脸配准,该方法不但能够筛选鲁棒的特征,而且可以将模型的大小压缩到原来的5%左右.接着,在某些关键点周围提取高维的纹理特征,并通过稀疏投影降维.对于服务器端,采用级联形状和纹理特征的方式进行高效的相似脸检索.首先基于稀疏形状重构的方式筛选脸型相似的人脸,然后基于稀疏纹理重构的方法确定相似脸.在三星Note 3智能手机上,人脸图像的配准时间约10 ms.在扩展的LFW(Labeled Face in Wild)数据库上,相似脸检索时间约1.5 s,整个模型大小约5.4 MB.大量实验结果表明,配准方法精度高,速度快,模型小;相似脸检索的方法效率高,检索结果更符合人们的视觉感受.展开更多
基金supported by China Postdoctoral Science Foundation(2015M582355)the Doctor Scientific Research Start Project from Hubei University of Science and Technology(BK1418)National Natural Science Foundation of China(61271256)
文摘Face recognition systems have been in the active research in the area of image processing for quite a long time. Evaluating the face recognition system was carried out with various types of algorithms used for extracting the features, their classification and matching. Similarity measure or distance measure is also an important factor in assessing the quality of a face recognition system. There are various distance measures in literature which are widely used in this area. In this work, a new class of similarity measure based on the Lp metric between fuzzy sets is proposed which gives better results when compared to the existing distance measures in the area with Linear Discriminant Analysis (LDA). The result points to a positive direction that with the existing feature extraction methods itself the results can be improved if the similarity measure in the matching part is efficient.
文摘This paper presents a method which utilizes color, local symmetry and geometry information of human face based on various models. The algorithm first detects most likely face regions or ROIs (Region-Of-Interest) from the image using face color model and face outline model, produces a face color similarity map. Then it performs local symmetry detection within these ROIs to obtain a local symmetry similarity map. The two maps and local similarity map are fused to obtain potential facial feature points. Finally similarity matching is performed to identify faces between the fusion map and face geometry model under affine transformation. The output results are the detected faces with confidence values. The experimental results demonstrate its validity and robustness to identify faces under certain variations.
文摘为解决目前深度仿造检测方法对于跨数据集的检测性能难以提高的问题,提出基于注意力机制和一致性损失相结合的深度伪造人脸检测方法(method based on attention mechanism and consistency loss,MAMCL)。采用多注意力机制,迫使网络捕捉到更细微的局部异常。采用基于注意力机制的擦除方式,鼓励模型深入挖掘之前忽略的区域。设计一致性模块获取伪造图像中普遍存在的不一致细节特征,并应用一致性损失引导模型更加关注伪造细节。在面部取证++(FaceForensics++,FF++)数据集上进行实验,准确率达到96.38%,受试者工作特征曲线(receiver operating characteristic curve,ROC)的曲线下面积达到99.34%,在泛化性能测试中也取得了良好的效果。通过消融实验,证明了每个模块的有效性。结果表明,提出的检测方法能够较为准确地检测深度伪造人脸,且具有良好的泛化性能,可以作为应对当前人脸伪造威胁的有效检测手段。
文摘近年来,稀疏表示分类(Sparse Representation Based Classification,SRC)方法在人脸识别中受到越来越多的关注。原始SRC方法使用所有的训练样本组成字典矩阵,当训练样本比较多时,稀疏系数的求解会变得非常耗时。为了解决这一问题,提出一种新的局部稀疏表示分类(Local SRC,LSRC)方法。该方法针对每个测试样本,根据测试样本和训练样本稀疏系数之间的相似性来选择部分训练样本,由这些训练样本组成字典,然后在这个字典上对测试样本进行稀疏分解。该方法性能相比于原始LSRC方法更稳定。在ORL、Yale和AR人脸库上的实验结果表明,该方法的效果优于SRC和LSRC。
文摘主要研究了移动平台上的相似脸检索问题.对于移动端,首先采用基于稀疏约束的级联回归模型进行精确的人脸配准,该方法不但能够筛选鲁棒的特征,而且可以将模型的大小压缩到原来的5%左右.接着,在某些关键点周围提取高维的纹理特征,并通过稀疏投影降维.对于服务器端,采用级联形状和纹理特征的方式进行高效的相似脸检索.首先基于稀疏形状重构的方式筛选脸型相似的人脸,然后基于稀疏纹理重构的方法确定相似脸.在三星Note 3智能手机上,人脸图像的配准时间约10 ms.在扩展的LFW(Labeled Face in Wild)数据库上,相似脸检索时间约1.5 s,整个模型大小约5.4 MB.大量实验结果表明,配准方法精度高,速度快,模型小;相似脸检索的方法效率高,检索结果更符合人们的视觉感受.
