异常检测是模式识别领域的经典研究,然而在极端类别不平衡场景下,异常样本匮乏,训练数据仅包含正常样本,传统异常检测方法难以适用。因此,单分类算法逐渐受到关注,它只使用目标类样本构建决策边界,实现对非目标类样本的识别。目前单分...异常检测是模式识别领域的经典研究,然而在极端类别不平衡场景下,异常样本匮乏,训练数据仅包含正常样本,传统异常检测方法难以适用。因此,单分类算法逐渐受到关注,它只使用目标类样本构建决策边界,实现对非目标类样本的识别。目前单分类算法已经取得了显著进展,然而也存在一些局限性:(1)原始特征空间容易受噪声特征干扰;(2)单模型的单分类算法难以从多个特征空间学习更全面的决策边界;(3)缺少对先前模型的欠拟合样本进行针对性学习。为了解决这些问题,本文提出了基于宽度自编码网络的单分类集成算法(Ensemble One-class Classification Based on BLS-Autoencoder,EOC-BLSAE)。首先,本文设计了一种单分类宽度自编码网络模型(One-class BLS-Autoencoder,OC-BLSAE),它能高效学习原始特征空间到重构特征空间的非线性映射关系,利用重构误差构建决策边界;接着,本文设计了单分类Boosting策略,通过最小化全局重构损失,迭代学习欠拟合样本,从而多角度构建OC-BLSAE模型,并自适应评估模型的可靠性;最终,加权集成多个OC-BLSAE模型,有效提升整体算法准确性和鲁棒性。在实验中,本文在16个不同规模的单分类任务上进行参数实验、对比实验和消融实验,结果表明所提算法参数选择较为灵活,算法各模块能够相互协同,有效提升单分类任务的准确性和鲁棒性,整体性能超过前沿单分类方法。展开更多
Visible-infrared person re-identification(VIPR), is a cross-modal retrieval task that searches a target from a gallery captured by cameras of different spectrums.The severe challenge for VIPR is the large intra-class ...Visible-infrared person re-identification(VIPR), is a cross-modal retrieval task that searches a target from a gallery captured by cameras of different spectrums.The severe challenge for VIPR is the large intra-class variation caused by the modal discrepancy between visible and infrared images.For that, this paper proposes a query related cluster(QRC) method for VIPR.Firstly, this paper uses an attention mechanism to calculate the similarity relation between a visible query and infrared images with the same identity in the gallery.Secondly, those infrared images with the same query images are aggregated by using the similarity relation to form a dynamic clustering center corresponding to the query image.Thirdly, QRC loss function is designed to enlarge the similarity between the query image and its dynamic cluster center to achieve query related clustering, so as to compact the intra-class variations.Consequently, in the proposed QRC method, each query has its own dynamic clustering center, which can well characterize intra-class variations in VIPR.Experimental results demonstrate that the proposed QRC method is superior to many state-of-the-art approaches, acquiring a 90.77% rank-1 identification rate on the RegDB dataset.展开更多
Fabric retrieval is very challenging since problems like viewpoint variations,illumination changes,blots,and poor image qualities are usually encountered in fabric images.In this work,a novel deep feature nonlinear fu...Fabric retrieval is very challenging since problems like viewpoint variations,illumination changes,blots,and poor image qualities are usually encountered in fabric images.In this work,a novel deep feature nonlinear fusion network(DFNFN)is proposed to nonlinearly fuse features learned from RGB and texture images for improving fabric retrieval.Texture images are obtained by using local binary pattern texture(LBP-Texture)features to describe RGB fabric images.The DFNFN firstly applies two feature learning branches to deal with RGB images and the corresponding LBP-Texture images simultaneously.Each branch contains the same convolutional neural network(CNN)architecture but independently learning parameters.Then,a nonlinear fusion module(NFM)is designed to concatenate the features produced by the two branches and nonlinearly fuse the concatenated features via a convolutional layer followed with a rectified linear unit(ReLU).The NFM is flexible since it can be embedded in different depths of the DFNFN to find the best fusion position.Consequently,DFNFN can optimally fuse features learned from RGB and LBP-Texture images to boost the retrieval accuracy.Extensive experiments on the Fabric 1.0 dataset show that the proposed method is superior to many state-of-the-art methods.展开更多
近年来,作为一种能够提供更富有沉浸感的多媒体媒质,光场图像(Light Field Image,LFI)引起广泛的关注。针对光场图像数据量巨大的问题,本文提出了一种基于多视点伪序列的光场图像高效压缩方案。在编码端,所提方法首先将光场相机捕获得...近年来,作为一种能够提供更富有沉浸感的多媒体媒质,光场图像(Light Field Image,LFI)引起广泛的关注。针对光场图像数据量巨大的问题,本文提出了一种基于多视点伪序列的光场图像高效压缩方案。在编码端,所提方法首先将光场相机捕获得到的原始光场图像根据相机的微透镜阵列分解成子孔径图像。接着根据子孔径图像存在较强视点内和视点间相关性,选取部分子孔径图像进行多视点伪序列构建,基于MV-HEVC设计适用于多视点伪序列的预测编码结构进行编码。在解码端,所提方法基于已解码多视点伪序列通过视频帧插值方法重建出未编码传输的子孔径视图,从而重建出全部光场图像。实验结果表明本文所提算法优于现有基于视差引导稀疏编码的光场图像压缩方法,BD-rate平均节约18.5%,BD-PSNR平均提高1.28 dB。展开更多
文摘异常检测是模式识别领域的经典研究,然而在极端类别不平衡场景下,异常样本匮乏,训练数据仅包含正常样本,传统异常检测方法难以适用。因此,单分类算法逐渐受到关注,它只使用目标类样本构建决策边界,实现对非目标类样本的识别。目前单分类算法已经取得了显著进展,然而也存在一些局限性:(1)原始特征空间容易受噪声特征干扰;(2)单模型的单分类算法难以从多个特征空间学习更全面的决策边界;(3)缺少对先前模型的欠拟合样本进行针对性学习。为了解决这些问题,本文提出了基于宽度自编码网络的单分类集成算法(Ensemble One-class Classification Based on BLS-Autoencoder,EOC-BLSAE)。首先,本文设计了一种单分类宽度自编码网络模型(One-class BLS-Autoencoder,OC-BLSAE),它能高效学习原始特征空间到重构特征空间的非线性映射关系,利用重构误差构建决策边界;接着,本文设计了单分类Boosting策略,通过最小化全局重构损失,迭代学习欠拟合样本,从而多角度构建OC-BLSAE模型,并自适应评估模型的可靠性;最终,加权集成多个OC-BLSAE模型,有效提升整体算法准确性和鲁棒性。在实验中,本文在16个不同规模的单分类任务上进行参数实验、对比实验和消融实验,结果表明所提算法参数选择较为灵活,算法各模块能够相互协同,有效提升单分类任务的准确性和鲁棒性,整体性能超过前沿单分类方法。
基金Supported by the National Natural Science Foundation of China (No.61976098)the Natural Science Foundation for Outstanding Young Scholars of Fujian Province (No.2022J06023)。
文摘Visible-infrared person re-identification(VIPR), is a cross-modal retrieval task that searches a target from a gallery captured by cameras of different spectrums.The severe challenge for VIPR is the large intra-class variation caused by the modal discrepancy between visible and infrared images.For that, this paper proposes a query related cluster(QRC) method for VIPR.Firstly, this paper uses an attention mechanism to calculate the similarity relation between a visible query and infrared images with the same identity in the gallery.Secondly, those infrared images with the same query images are aggregated by using the similarity relation to form a dynamic clustering center corresponding to the query image.Thirdly, QRC loss function is designed to enlarge the similarity between the query image and its dynamic cluster center to achieve query related clustering, so as to compact the intra-class variations.Consequently, in the proposed QRC method, each query has its own dynamic clustering center, which can well characterize intra-class variations in VIPR.Experimental results demonstrate that the proposed QRC method is superior to many state-of-the-art approaches, acquiring a 90.77% rank-1 identification rate on the RegDB dataset.
基金the National Natural Science Foundation of China(No.61976098,61871434,61802136,61876178)Open Foundation of Key Laboratory of Security Prevention Technology and Risk Assessment,People’s Public Security University of China(No.18AFKF11)+1 种基金Science and Technology Bureau of Quanzhou(No.2018C115R)the Subsidized Project for Postgraduates’Innovative Fund in Scientific Research of Huaqiao University(No.18014084008).
文摘Fabric retrieval is very challenging since problems like viewpoint variations,illumination changes,blots,and poor image qualities are usually encountered in fabric images.In this work,a novel deep feature nonlinear fusion network(DFNFN)is proposed to nonlinearly fuse features learned from RGB and texture images for improving fabric retrieval.Texture images are obtained by using local binary pattern texture(LBP-Texture)features to describe RGB fabric images.The DFNFN firstly applies two feature learning branches to deal with RGB images and the corresponding LBP-Texture images simultaneously.Each branch contains the same convolutional neural network(CNN)architecture but independently learning parameters.Then,a nonlinear fusion module(NFM)is designed to concatenate the features produced by the two branches and nonlinearly fuse the concatenated features via a convolutional layer followed with a rectified linear unit(ReLU).The NFM is flexible since it can be embedded in different depths of the DFNFN to find the best fusion position.Consequently,DFNFN can optimally fuse features learned from RGB and LBP-Texture images to boost the retrieval accuracy.Extensive experiments on the Fabric 1.0 dataset show that the proposed method is superior to many state-of-the-art methods.
文摘本文提出一种采用尺度不变特征变换(Scale-Invariant Feature Transform,SIFT)和局部聚合向量(Vector of Locally Aggregated Descriptors,VLAD)特征编码的布匹检索算法。首先,提取图像的SIFT特征,以对图像进行特征表达。但是,每张图像SIFT特征点数量可能不同,导致不同图像的特征向量维度不一致,无法直接进行图像之间的相似度计算。为此,本文进一步对图像的SIFT特征进行VLAD编码,在保证不同图像的特征维度一致的同时,改进SIFT特征对图像的表达能力。在VLAD编码方面,先用K-means聚类算法生成视觉词典;再进行特征向量局部聚合。局部聚合过程包括:首先,计算图像中SIFT特征向量与对应视觉词之间的残差;然后,将每个视觉词相应的残差求和;最后,把各个视觉词上的残差求和值进行串联得到图像的VLAD编码。本文实验采用十次平均的累计匹配特性(Cumulative Match Characteristic,CMC)曲线作为性能指标。结果表明,本文所提出的方法能提高检索速度,且具有较高的识别率,其平均Rank 1识别率达到95.03%。
文摘近年来,作为一种能够提供更富有沉浸感的多媒体媒质,光场图像(Light Field Image,LFI)引起广泛的关注。针对光场图像数据量巨大的问题,本文提出了一种基于多视点伪序列的光场图像高效压缩方案。在编码端,所提方法首先将光场相机捕获得到的原始光场图像根据相机的微透镜阵列分解成子孔径图像。接着根据子孔径图像存在较强视点内和视点间相关性,选取部分子孔径图像进行多视点伪序列构建,基于MV-HEVC设计适用于多视点伪序列的预测编码结构进行编码。在解码端,所提方法基于已解码多视点伪序列通过视频帧插值方法重建出未编码传输的子孔径视图,从而重建出全部光场图像。实验结果表明本文所提算法优于现有基于视差引导稀疏编码的光场图像压缩方法,BD-rate平均节约18.5%,BD-PSNR平均提高1.28 dB。
