Hierarchical Sparse Spectral Clustering For Image Set Classification

In this paper, we propose a simultaneous feature and dictionary learning (SFDL) method for image set based face recognition, where each training and testing example contains a face image set captured from different poses, illuminations, expressions and resolutions. While several feature learning and dictionary learning methods have been proposed for image set based face recognition in recent years, most of them learn the features and dictionaries separately, which may not be powerful enough because some discriminative information for dictionary learning may be compromised in the feature learning stage if they are applied sequentially, and vice versa. To address this, we propose a SFDL method to learn discriminative features and dictionaries simultaneously from raw face images so that discriminative information can be jointly exploited. Extensive experimental results on four widely used face datasets show that our method achieves better performance than state-of-the-art image set based face recognition methods.KeywordsFace recognitionimage setfeature learningdictionary learningsimultaneous learning

We present a no-reference (NR) image quality assessment (IQA) algorithm that is inspired by the representation of visual scenes in the primary visual cortex of the human visual system. Specifically, we use the sparse coding model of the area V1 to construct an overcomplete dictionary for sparsely representing pristine (undistorted) natural images. First, we empirically demonstrate that the distribution of the sparse representation coefficients of natural images have sharp peaks and heavy tails, and can therefore be modeled using a Univariate Generalized Gaussian Distribution (UGGD). We then show that the UGGD model parameters form good features for distortion estimation and formulate our no-reference IQA algorithm based on this observation. Subsequently, we find UGGD model parameters that are representative of the class of pristine natural images. This is achieved using a training set of undistorted natural images. The perceptual quality of a test image is then defined to be the likelihood of its sparse coefficients being generated from the pristine UGGD model. We show that the proposed algorithm correlates well with subjective evaluation over several standard image databases. Further, the proposed method allows us to construct a distortion map that has several useful applications like distortion localization, adaptive rate allocation etc. Finally and importantly, the proposed NR-IQA algorithm does not make use of any distortion information or subjective scores during the training process.

Image set based face recognition has recently become a popular topic as it has better performance than single image based face recognition. However, preprocessing is needed to remove the effects of some adverse conditions such as different pose angles, illumination, and expression differences within the set. One of the most effective preprocessing to improve the face recognition rate is face frontalization. Face frontalization is defined as the artificial acquisition of a face image with a different pose angle to a frontal pose. It has been observed that this process increases the face recognition performance. In this paper, image set based face recognition was performed by applying face frontalization to all images in the sets. Firstly, the faces in the IJBA database were frontalized by using the Rotate and Render hybrid frontalization method, which is based on a Three-Dimensional and Generative Adversial Network. Then, discriminative convex classifier is used for set based face recognition. In face recognition experiments, when the frontalized IJBA database and its non-frontalized version were compared, it was observed that the accuracy of face recognition increased with the frontalized face images.

Based on the linear unmixing images of different surface objects, online dictionary learning algorithm was utilized to compute the sparse dictionaries for multispectral linear unmixing images and panchromatic images. Principal component analysis (PCA) was then utilized to generate united sparse PCA dictionaries through the extraction of the first principal components of panchromatic images and unmixing image dictionaries. The number of dictionaries is determined to be 480 after taking into consideration of the limitation in computing power and root-mean-square error of restructured images. Based on these dictionaries, orthogonal matching pursuit method was utilized to calculate the sparse coefficients of multispectral and panchromatic images, separately, while nonnegative matrix factorization fusion algorithm was utilized to calculate multispectral and panchromatic sparse coefficients to obtain sparse coefficient of the fusional image on all bands, with the resulted matrix having a size of 480 × 255 025. These united sparse PCA dictionaries and fusion sparse coefficients were then used to reconstruct the fusional image. Through the analysis of five quantitative indices of fusion assessment, the proposed fusion algorithm has retained the multispectral information of images and enhanced the detailed information in image texture.

Majority of the image set based face recognition methods use a generatively learned model for each person that is learned independently by ignoring the other persons in the gallery set. In contrast to these methods, this paper introduces a novel method that searches for discriminative convex models that best fit to an individual's face images but at the same time are as far as possible from the images of other persons in the gallery. We learn discriminative convex models for both affine and convex hulls of image sets. During testing, distances from the query set images to these models are computed efficiently by using simple matrix multiplications, and the query set is assigned to the person in the gallery whose image set is closest to the query images. The proposed method significantly outperforms other methods using generative convex models in terms of both accuracy and testing time, and achieves the state-of-the-art results on four of the five tested datasets. Especially, the accuracy improvement is significant on the challenging PaSC, COX and ESOGU video datasets.

Image fusion is the process which aims to integrate the relevant and complementary information from a set of images into a single comprehensive image. Sparse representation (SR) is a powerful technique used in a wide variety of applications like denoising, compression and fusion. Building a compact and informative dictionary is the principal challenge in these applications. Hence, we propose a supervised classification based learning technique for the fusion algorithm. As an initial step, each patch of the training data set is pre-classified based on their gradient dominant direction. Then, a dictionary is learned using K-SVD algorithm. With this universal dictionary, sparse coefficients are estimated using greedy OMP algorithm to represent the given set of source images in the dominant direction. Finally, the Euclidean norm is used as a distance measure to reconstruct the fused image. Experimental results on different types of source images demonstrate the effectiveness of the proposed algorithm with conventional methods in terms of visual and quantitative evaluations.

Clustering large hyperspectral images (HSIs) is a very challenging problem because large HSIs have high dimensionality, large spectral variability, and large computational and memory consumption. Recently, sparse subspace clustering (SSC) has achieved remarkable success in HSI clustering. However, most SSC-based methods suffer from the following bottlenecks for large HSIs: 1) high computational consumption and memory space during the construction of the similarity matrix and decomposition of the graph Laplacian matrix and 2) failure to capture the relationships among dictionary atoms, sparse coefficients, and hyperspectral pixels. To address these challenges, we propose a novel algorithm that extends SSC to cocluster large HSIs, called bipartite graph partition with graph nonnegative matrix factorization (BGP-GNMF). Specifically, to fully explore the characteristics of the spectral and spatial contexts in HSIs, we propose a novel superpixel and pixel coclustering framework with bipartite graph partitioning in the joint sparse representation domain, where superpixel-based dictionary atoms are defined as disjoint vertex sets of the bipartite graph and the joint sparsity representation is mapped into the adjacency matrix of the undirected bipartite graph. To overcome the challenges of high computational consumption and large memory space for large HSIs, the bipartite graph partition with orthonormal constrained nonnegative matrix factorization is proposed to simultaneously cluster the structured dictionary atoms and hyperspectral pixels with an indicator matrix. Finally, to exploit the intrinsic geometry of HSIs, we incorporate manifold regularization into the bipartite graph partition to improve final clustering accuracy. The effectiveness and efficiency of the proposed method are verified on three classical HSIs, and the experimental results illustrate the superiority of the proposed method compared with other state-of-the-art HSI clustering methods.

Sparse representation (SR) has shown great potential in multishot person reidentification. However, conventional SR methods usually encode each probe image individually, which ignores the prior knowledge shared by similar probe images with the same identity. Considering that probe images with the same identity are similar and can be complementary to each other, we propose a joint group sparse representation (JGSR) method to simultaneously encode a set of probe person images with the same identity. Additionally, we consider the fact that prior knowledge is also contained in the set of images with the same identity in the gallery. In JGSR, probe images with the same identity are coded on the same dictionary atoms and shared similar coefficients. Furthermore, the person reidentification performance is also improved by incorporating JGSR with k-means clustering and kernel local Fisher discriminant analysis into a unified framework. Extensive experiments on three publicly available iLIDS-VID, PRID 2011, and SAIVT-SoftBio multishot benchmark datasets were conducted and demonstrated the superior performance of the JGSR method in comparison with current state-of-the-art methods.

Fisher discrimination dictionary sparse learning (FDDL) has led to interesting image recognition results where the Fisher discrimination criterion is subject to the coding coefficients. But Fisher discrimination criterion has the limitations of data distribution assumptions and does not consider the local manifold structure of the coding coefficients. In this paper, we will introduce a novel Fisher discrimination sparse learning based on graph embedding (GE-FDSL) scheme. First, we utilizes graph embedding framework to define intra-class compact matrix and inter-class separable matrix imposed on the coding coefficients of training samples to preserving the intra-class compactness and the inter-class separability for the training samples, which simultaneously consider the local manifold structure and label information of the coding coefficients. Then, a new Fisher discrimination criterion based on graph embedding is added to the object function of the sparse coding problem so that the coding coefficients have more discriminative power, where the dictionary atoms in the sparse coding model are associated with the class labels so that the reconstructed error is applied to classification. This method can learn a structured dictionary and sparse coefficients, and in the meantime, it will also keep the local manifold structure of the coding coefficients. So, they will be more discriminative. Experiments on many image databases show that the our algorithm has good classification and recognition performance.

Sparse manifold learning algorithms combine techniques in manifold learning and sparse optimization to learn features that could be utilized for downstream tasks. The standard setting of compressive sensing can not be immediately applied to this setup. Due to the intrinsic geometric structure of data, dictionary atoms might be redundant and do not satisfy the restricted isometry property or coherence condition. In addition, manifold learning emphasizes learning local geometry which is not reflected in a standard ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> minimization problem. We propose weighted ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf> and weighted ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> regularizations that encourage representation via neighborhood atoms suited for dictionary based manifold learning. Assuming that the data is generated from Delaunay triangulation, we show the equivalence of weighted ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> and weighted ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf> . We discuss an optimization program that learns the dictionaries and sparse co-efficients and demonstrate the utility of our regularization on synthetic and real datasets.

ℓ1-K-SVD: A robust dictionary learning algorithm with simultaneous update

Bearing faults are common issues in industrial equipment. Traditional fault diagnosis methods primarily rely on manually crafted features and require prior knowledge. Dictionary learning is a adaptive method that can automatically extract features from signals, reduce the dependence on prior knowledge. However, dictionary learning is often considered a black-box approach, which lacks interpretability. This paper confirms that high-frequency impulsive signals and characteristic fault frequencies can be captured through time-domain and envelope analysis of the learned atoms. Experimental results show that dictionary atoms more frequently capture high-frequency impulsive components, underscoring their importance in fault diagnosis. Moreover, the sparse coefficients corresponding to different faults are significantly different and can be effectively used for classification tasks. Further analysis of the experimental results enhances the interpretability of dictionary learning in bearing fault diagnosis.

The nonparametric Bayesian dictionary learning based interpolation method for WSNs missing data

Spectral clustering (SC) is one of the most popular clustering methods and often outperforms traditional clustering methods. SC uses the eigenvectors of a Laplacian matrix calculated from a similarity matrix of a dataset. SC has serious drawbacks: the significant increases in the time complexity derived from the computation of eigenvectors and the memory space complexity to store the similarity matrix. To address the issues, I develop a new approximate spectral clustering using the network generated by growing neural gas (GNG), called ASC with GNG in this study. ASC with GNG uses not only reference vectors for vector quantization but also the topology of the network for extraction of the topological relationship between data points in a dataset. ASC with GNG calculates the similarity matrix from both the reference vectors and the topology of the network generated by GNG. Using the network generated from a dataset by GNG, ASC with GNG achieves to reduce the computational and space complexities and improve clustering quality. In this study, I demonstrate that ASC with GNG effectively reduces the computational time. Moreover, this study shows that ASC with GNG provides equal to or better clustering performance than SC.

Due to the serious and complicated noise in the shallow sea environment, the received signal obtained by the hydrophone is disturbed by the noise to a large extent. It has a low signal-to-noise ratio (SNR), which leads to problems such as difficulty in processing the underwater acoustic signal. To solve this problem, to more effectively remove the ocean noise in the useful signal, a denoising method based on sparse decomposition and dictionary learning is adopted. First, a complete Discrete Cosine Transform (DCT) dictionary is randomly constructed. Then the orthogonal matching pursuit (OMP) is used to represent the noisy underwater acoustic signal sparsely, the method of optimal directions (MOD) and K-singular value decomposition algorithm (K-SVD) are used to update the complete DCT dictionary respectively. According to the updated new dictionary and sparse coefficients, the underwater acoustic signal is reconstructed, and the ocean noise is removed. By denoising the different form of simulated signals with different SNRs, the results show that two methods can remove various noises mixed in the underwater acoustic signal effectively and retain the signal details while denoising. The SNR gain can reach about 20dB.