As one of the most popular dimensionality reduction techniques, locality preserving projections (LPP) has been widely used in computer vision and pattern recognition. However, practical applications, data is always corrupted by noises. For data, samples from same class may not be distributed nearest area, thus LPP lose its effectiveness. In this paper, it assumed that grossly noise matrix sparse. Based on these assumptions, we propose a novel method, named low-rank (LRPP) for image...

Robustness to noises, outliers, and corruptions is an important issue in linear dimensionality reduction. Since the sample-specific outliers exist, class-special structure or local geometric destroyed, thus, many existing methods, including popular manifold learning- based fail achieve good performance recognition tasks. In this paper, we focus on unsupervised robust reduction corrupted data by introducing low-rank representation (LRR). Thus, a technique termed embedding (LRE) proposed which...

As a popular dimensionality reduction method, nonnegative matrix factorization (NMF) has been widely used in image classification. However, the NMF does not consider discriminant information from data themselves. In addition, most NMF-based methods use Euclidean distance as metric, which is sensitive to noise or outliers data. To solve these problems, this paper, we introduce structural incoherence and low-rank propose novel called structurally incoherent (SILR-NMF), jointly properties of...

2-D neighborhood preserving projection (2DNPP) uses images as feature input instead of 1-D vectors used by (NPP). 2DNPP requires less computation time than NPP. However, both NPP and use the L 2 norm a metric, which is sensitive to noise in data. In this paper, we proposed novel method called low-rank (LR-2DNPP). This divided data into component part that encoded features, an error ensured was sparse. Then, nearest neighbor graph learned from clean using same procedure 2DNPP. To ensure...

As one of the most prevalent branches transfer learning, domain adaptation is dedicated to generalizing knowledge a source target perform machine learning tasks. In adaptation, key strategy overcome shift between different domains and learn shared features with invariance. However, existing methods focus on extracting common domains, do not consider problem class center in caused by this process. Specifically, when we align distributions, often ignore inherent feature attributes data, or...

As a branch of transfer learning, domain adaptation leverages useful knowledge from source to target for solving tasks. Most the existing methods focus on how diminish conditional distribution shift and learn invariant features between different domains. However, two important factors are overlooked by most methods: 1) transferred should be not only but also discriminative correlated, 2) negative avoided as much possible To fully consider these in adaptation, we propose guided discrimination...

Nonnegative matrix factorization (NMF), which aims at obtaining the nonnegative low-dimensional representation of data, has received wide attention. To obtain more effective discriminant bases from original NMF, in this paper, a novel method called (NDMF) is proposed for image classification. NDMF integrates constraint, orthogonality, and information objective function. considers incoherent both factors standard NMF to enhance ability learned base matrix. projects subspace regularize...

Domain adaptation leverages rich knowledge from a related source domain so that it can be used to perform tasks in target domain. For more obtained under relaxed conditions, methods have been widely pattern recognition and image classification. However, most of the existing only consider how minimize different distributions domains, which neglects what should transferred for specific task suffers negative transfer by distribution outliers. To address these problems, this paper, we propose...

In transfer learning, how to effectively useful information from the source domain target is crucial. this paper, we propose a novel learning method for image classification, named manifold via discriminant regression analysis (MTL-DRA), local geometry structure and ensure that transform matrix robust or sparse so samples different domains can be well combined. MTL-DRA, encode of by introducing between- within-class graphs preserve similarity reduce between-class similarity. With norms as...

Two-dimensional locality preserving projections (2DLPP) that use 2D image representation in projection learning can preserve the intrinsic manifold structure and local information of data. However, 2DLPP is based on Euclidean distance, which sensitive to noise outliers In this paper, we propose a novel method called nuclear norm-based two-dimensional (NN-2DLPP). First, NN-2DLPP recovers noisy data matrix through low-rank learning. Second, removed learned clean points are projected new...

In unsupervised domain adaptation (UDA), negative transfer is one of the most challenging problems. Due to complex environments, used data are always corrupted by noise or outliers in many applications. If noisy directly for adaptation, disturbances and influence also shifted target tasks. Thus, preventing effects caused key problems UDA that need be addressed. this article, a low-rank correlation learning (LRCL) method proposed UDA. LRCL, recovered learning; then both cleaned. Hence,...

2-D linear discriminant analysis (2DLDA) has been widely used in pattern recognition and image classification. 2DLDA selects discriminative features from the up left corner of images. However, uses Frobenius norm (F-norm), which is sensitive to noise or outliers data, as a metric. In this paper, we propose novel framework, called horizontal vertical nuclear norm-based (HVNN-2DLDA) for representation. proposed HVNN-2DLDA methods (i.e., HNN-2DLDA VNN-2DLDA) are proposed, both use criterion....

As a branch of domain adaptation (DA), multi-source DA (MSDA) is challenging issue that aims to transfer knowledge from multiple well-labeled source domains target for tasks. However, most existing related works focus on single-target adaptation, and not accounted for. We believe provide valuable knowledge. Meanwhile, in multi-target scenarios, feature generators with static parameters have difficulty generating deep features each individual domain. In this paper, we propose Dynamic...