Xianzhong Long

Discriminative graph regularized extreme learning machine and its application to face recognition

Extreme Learning Machine (ELM) has been proposed as a new algorithm for training single hidden layer feed forward neural networks. The main merit of ELM lies in the fact that the input weights as well as hidden layer bias are randomly generated and thus the output weights can be obtained analytically, which can overcome the drawbacks incurred by gradient-based training algorithms such as local optima, improper learning rate and low learning speed. Based on the consistency property of data, which enforces similar samples to share similar properties, we propose a discriminative graph regularized Extreme Learning Machine (GELM) for further enhancing its classification performance in this paper. In the proposed GELM model, the label information of training samples are used to construct an adjacent graph and correspondingly the graph regularization term is formulated to constrain the output weights to learn similar outputs for samples from the same class. The proposed GELM model also has a closed form solution as the standard ELM and thus the output weights can be obtained efficiently. Experiments on several widely used face databases show that our proposed GELM can achieve much performance gain over standard ELM and regularized ELM. Moreover, GELM also performs well when compared with the state-of-the-art classification methods for face recognition.

Graph regularized discriminative non-negative matrix factorization for face recognition

Non-negative matrix factorization (NMF) has been widely employed in computer vision and pattern recognition fields since the learned bases can be interpreted as a natural parts-based representation of the input space, which is consistent with the psychological intuition of combining parts to form a whole. In this paper, we propose a novel constrained nonnegative matrix factorization algorithm, called the graph regularized discriminative non-negative matrix factorization (GDNMF), to incorporate into the NMF model both intrinsic geometrical structure and discriminative information which have been essentially ignored in prior works. Specifically, both the graph Laplacian and supervised label information are jointly utilized to learn the projection matrix in the new model. Further we provide the corresponding multiplicative update solutions for the optimization framework, together with the convergence proof. A series of experiments are conducted over several benchmark face datasets to demonstrate the efficacy of our proposed GDNMF.

Graph Based Semi-Supervised Learning via Structure Preserving Low-Rank Representation

Semi-supervised learning works on utilizing both labeled and unlabeled data to improve learning performance, which has been receiving increasing attention in many applications such as clustering and classification. In this paper, we focus on the semi-supervised learning methods developed on data graph whose edge weights are measured by low-rank representation (LRR) coefficients. Specifically, we impose two constraints on LRR when constructing the graph: local affinity and distant repulsion, to preserve the data manifold information. The proposed model, termed structure preserving LRR (SPLRR), can preserve the local geometrical structure and without distorting the distant repulsion property. Using the augmented Lagrange multiplier (ALM) method framework, we derive an efficient approach to optimizing the SPLRR model. Experiments are conducted on four widely used data sets to validate the effectiveness of our proposed SPLRR model and the results demonstrate that SPLRR is an excellent model for graph based semi-supervised learning in comparison with the state-of-the-art methods.

A Fragile Watermarking Scheme Based on Hash Function for Web Pages

With the development of Internet, the integrity protection of web pages has become an important problem. A fragile watermarking scheme based on hash function for web pages is proposed in this paper. Watermarks are generated by using hash function and these watermarks are related to the web pages content. Then, the watermarks are embedded into the tags and embedding locations, which are randomly pointed. In the detection phase, regenerated watermarks and extracted watermarks are got from the watermarked web page. Finally, according to comparing the mentioned above two kind watermarks, we can judge whether the web page has been tampered or not. In addition, we can also identify the types of modification and locate the locations of modification.

Image classification based on nearest neighbor basis vectors

Image classification can be roughly divided into two categories, i.e., scene recognition and object recognition. There are two important steps in object recognition: Dictionary Learning and Feature Coding. In order to get the best classification performance, the optimal dictionary learning method and feature coding strategy should be used simultaneously. However, researchers recently have found that feature coding was more important than dictionary learning when sparse coding scheme was employed. With a dictionary formed by a random sample of descriptors, satisfactory results were obtained. Inspired by the discovery, in this paper we propose an image classification method based on nearest neighbor basis vectors of the dictionary. Each descriptor of image is linearly represented by its several nearest neighbor basis vectors. We exploit the widely used Spatial Pyramid Matching model (SPM) in our paper and name our method Nearest Neighbor Basis Vectors Spatial Pyramid Matching (NNBVSPM). In the NNBVSPM, the dictionary is generated by standard k-means clustering algorithm and the feature is encoded by our soft inner product coding scheme. Experimental results on scene 15 dataset and uiuc sports event dataset show that the proposed scheme outperforms some state-of-the-art methods.

A Fragile Watermarking Scheme Based on SVD for Web Pages

With the development of the Internet, more and more people concern the protection of web pages. Fragile watermarking technique is an effective method for solving the tamper-proof of web pages. In this paper, a fragile watermarking scheme based on singular value decomposition (SVD) for web pages is proposed. First of all, a model utilizing the frequency of letters is used to denote a web page, and the web page is expressed as a matrix through this model. Afterwards, we take SVD transformation on the matrix and extract its singular values to generate watermarks. Finally, the watermarks are embedded into the letters of HTML tags by altering the upper and lower cases of letters, simultaneously, they are embedded into the embedding locations controlled by a secret key. In the detection procedure, whether a web page has been tampered or not can be judged effectively only by comparing the generated watermarks with the extracted watermarks. The experiment shows that the proposed scheme has stronger abilities to identify the types of modifications and to locate the locations of modifications.

Discriminative Dictionary Learning Based on Supervised Feature Selection for Image Classification

The bag-of-features based models are widely used for image classification. In these models, an image is represented as a set of visual words which come from a dictionary. Therefore, a well learned dictionary is responsible for the discriminative power of representations of images. Our observations show that the representation of an image carries rich underlying information of a dictionary, so we propose a novel method to learn a dictionary by analyzing histogram representations of images, called Discriminative Dictionary Learning based on Supervised Feature Selection for Image Classification (DFS). Instead of directly learning a dictionary from the feature space, we construct a discriminative and compact dictionary from a coarse dictionary. The supervised feature selection technique is brought into the analysis of histogram representation, which eventually leads to dictionary refinement. Experimental results on challenging databases (Caltech-101, Caltech-256) show that learned dictionaries works better for bag-of-features based models.

A Fragile Watermarking Scheme for Tamper-Proof of Web Pages

With the development of the Internet, getting informationthrough browsing web pages has become a popularway. How to protect the integrity of a web page is anurgent problem. As an effective method, a novel fragilewatermarking scheme for tamper-proof of web pages isproposed in this paper for solving the problem. In ourscheme, the watermarks that are related to the content ofthe original web page are generated based on the SHA-1algorithm, and then they are embedded into the embeddinglocations controlled by two secret keys. In the detectionprocedure, whether a web page has been tampered or notcan be judged effectively only by comparing the generatedwatermarks with the extracted watermarks. The experimentshows that the proposed scheme has stronger abilities tolocate the locations of modifications and to identify the typesof modifications.

Sparse Non-negative Matrix Factorization Based on Spatial Pyramid Matching for Face Recognition

The non-negative matrix factorization (NMF) is a part-Based image representation method which allows only additive combinations of non-negative basis components. NMF has been widely used as a dimensionality reduction technique to solve problems in computer vision and pattern recognition fields. The sparse representation and spatial information of image are also important, however, existing NMF methods do not take these two aspects into consideration simultaneously. In this paper, we propose a novel NMF method with spatial information for face recognition, which is called sparse non-negative matrix factorization Based on spatial pyramid matching (SNMFSPM). Experimental results on several benchmark databases show that the proposed scheme outperforms some classical methods.

A Fragile Software Watermarking for Tamper-Proof

A fragile software watermarking scheme for integrity verification of software is proposed in this paper. The algorithm uses the idea of semantic-preserving code substitution for embedding the watermark. With the system, the generated watermark relates closely to the content of software, and the scheme has highly sensitive to different types of attacks. Furthermore, both watermark generating controlled by a key and watermark embedding position under the control of another key enhance the security of watermarking scheme. The scheme not only can effectively detect tampering, but also has the ability to identify the type of tampering clearly.