Match!

Multiple Kernel Sparse Representations for Supervised and Unsupervised Learning

Published on May 9, 2014in IEEE Transactions on Image Processing6.79
· DOI :10.1109/TIP.2014.2322938
Jayaraman J. Thiagarajan16
Estimated H-index: 16
(ASU: Arizona State University),
Karthikeyan Natesan Ramamurthy14
Estimated H-index: 14
(ASU: Arizona State University),
Andreas Spanias29
Estimated H-index: 29
(ASU: Arizona State University)
Abstract
In complex visual recognition tasks, it is typical to adopt multiple descriptors, which describe different aspects of the images, for obtaining an improved recognition performance. Descriptors that have diverse forms can be fused into a unified feature space in a principled manner using kernel methods. Sparse models that generalize well to the test data can be learned in the unified kernel space, and appropriate constraints can be incorporated for application in supervised and unsupervised learning. In this paper, we propose to perform sparse coding and dictionary learning in the multiple kernel space, where the weights of the ensemble kernel are tuned based on graph-embedding principles such that class discrimination is maximized. In our proposed algorithm, dictionaries are inferred using multiple levels of 1D subspace clustering in the kernel space, and the sparse codes are obtained using a simple levelwise pursuit scheme. Empirical results for object recognition and image clustering show that our algorithm outperforms existing sparse coding based approaches, and compares favorably to other state-of-the-art methods.
  • References (83)
  • Citations (57)
📖 Papers frequently viewed together
6,957 Citations
6,192 Citations
118 Citations
78% of Scinapse members use related papers. After signing in, all features are FREE.
References83
Newest
#1Hien Van Nguyen (Princeton University)H-Index: 17
#2Vishal M. Patel (UMD: University of Maryland, College Park)H-Index: 44
Last. Rama Chellappa (UMD: University of Maryland, College Park)H-Index: 98
view all 4 authors...
118 CitationsSource
May 26, 2013 in ICASSP (International Conference on Acoustics, Speech, and Signal Processing)
#1Yin Zhou (UD: University of Delaware)H-Index: 13
#2Jinglun Gao (UD: University of Delaware)H-Index: 3
Last. Kenneth E. Barner (UD: University of Delaware)H-Index: 31
view all 3 authors...
Discovering the intrinsic low-dimensional structure from high-dimensional observation space (e.g., images, videos), in many cases, is critical to successful recognition. However, many existing nonlinear manifold learning (NML) algorithms have quadratic or cubic complexity in the number of data, which makes these algorithms computationally exorbitant in processing real-world large-scale datasets. Randomly selecting a subset of data points is very likely to place NML algorithms at the risk of loca...
3 CitationsSource
May 26, 2013 in ICASSP (International Conference on Acoustics, Speech, and Signal Processing)
#1Rushil Anirudh (ASU: Arizona State University)H-Index: 8
#2Karthikeyan Natesan Ramamurthy (ASU: Arizona State University)H-Index: 14
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 5 authors...
In this paper, we consider low-dimensional and sparse representation models for human actions, that are consistent with how actions evolve in high-dimensional feature spaces. We first show that human actions can be well approximated by piecewise linear structures in the feature space. Based on this, we propose a new dictionary model that considers each atom in the dictionary to be an affine subspace defined by a point and a corresponding line. When compared to centered clustering approaches such...
3 CitationsSource
#1Jayaraman J. Thiagarajan (ASU: Arizona State University)H-Index: 16
#2Karthikeyan Natesan Ramamurthy (ASU: Arizona State University)H-Index: 14
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 3 authors...
Dictionaries adapted to the data provide superior performance when compared to predefined dictionaries in applications involving sparse representations. Algorithmic stability and generalization are desirable characteristics for dictionary learning algorithms that aim to build global dictionaries which can efficiently model any test data similar to the training samples. In this paper, we propose an algorithm to learn dictionaries for sparse representation of image patches, and prove that the prop...
8 Citations
#1Jayaraman J. Thiagarajan (ASU: Arizona State University)H-Index: 16
#2Karthikeyan Natesan Ramamurthy (ASU: Arizona State University)H-Index: 14
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 3 authors...
Mixing matrix estimation in instantaneous blind source separation (BSS) can be performed by exploiting the sparsity and disjoint orthogonality of source signals. As a result, approaches for estimating the unknown mixing process typically employ clustering algorithms on the mixtures in a parametric domain, where the signals can be sparsely represented. In this paper, we propose two algorithms to perform discriminative clustering of the mixture signals for estimating the mixing matrix. For the cas...
39 CitationsSource
Nov 11, 2012 in BIBE (Bioinformatics and Bioengineering)
#1Jayaraman J. Thiagarajan (ASU: Arizona State University)H-Index: 16
#2Deepta Rajan (ASU: Arizona State University)H-Index: 5
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 5 authors...
In this paper, we describe a pixel based approach for automated segmentation of tumor components from MR images. Sparse coding with data-adapted dictionaries has been successfully employed in several image recovery and vision problems. Since it is trivial to obtain sparse codes for pixel values, we propose to consider their non-linear similarities to perform kernel sparse coding in a high dimensional feature space. We develop the kernel K-lines clustering procedure for inferring kernel dictionar...
4 CitationsSource
#1Karthikeyan Natesan Ramamurthy (ASU: Arizona State University)H-Index: 14
#2Jayaraman J. Thiagarajan (ASU: Arizona State University)H-Index: 16
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 4 authors...
Several supervised, semi-supervised, and unsupervised machine learning schemes can be unified under the general framework of graph embedding. Incorporating graph embedding principles into sparse representation based learning schemes can provide an improved performance in several learning tasks. In this work, we propose a dictionary learning procedure for computing discriminative sparse codes that obey graph embedding constraints. In order to compute the graph-embedded sparse codes, we integrate ...
11 CitationsSource
Oct 7, 2012 in ECCV (European Conference on Computer Vision)
#1Hanlin Goh (Institute for Infocomm Research Singapore)H-Index: 10
#2Nicolas Thome (University of Paris)H-Index: 24
Last. Joo-Hwee Lim (Institute for Infocomm Research Singapore)H-Index: 26
view all 4 authors...
Recently, the coding of local features (e.g. SIFT) for image categorization tasks has been extensively studied. Incorporated within the Bag of Words (BoW) framework, these techniques optimize the projection of local features into the visual codebook, leading to state-of-the-art performances in many benchmark datasets. In this work, we propose a novel visual codebook learning approach using the restricted Boltzmann machine (RBM) as our generative model. Our contribution is three-fold. Firstly, we...
54 CitationsSource
Sep 1, 2012 in ICIP (International Conference on Image Processing)
#1Jayaraman J. Thiagarajan (ASU: Arizona State University)H-Index: 16
#2Karthikeyan Natesan Ramamurthy (ASU: Arizona State University)H-Index: 14
Last. Andreas Spanias (ASU: Arizona State University)H-Index: 29
view all 4 authors...
The success of sparse representations in image modeling and recovery has motivated its use in computer vision applications. Image retrieval and classification tasks require extracting features that discriminate different image classes. State-of-the-art object recognition methods based on sparse coding use spatial pyramid features obtained from dense descriptors. In this paper, we develop a feature extraction method that uses multiple global/local features extracted from large overlapping regions...
18 CitationsSource
Aug 12, 2012 in KDD (Knowledge Discovery and Data Mining)
#1Ashesh Jain (IITD: Indian Institute of Technology Delhi)H-Index: 1
#2S. V. N. Vishwanathan (Purdue University)H-Index: 35
Last. Manik Varma (Microsoft)H-Index: 24
view all 3 authors...
Multiple Kernel Learning (MKL) aims to learn the kernel in an SVM from training data. Many MKL formulations have been proposed and some have proved effective in certain applications. Nevertheless, as MKL is a nascent field, many more formulations need to be developed to generalize across domains and meet the challenges of real world applications. However, each MKL formulation typically necessitates the development of a specialized optimization algorithm. The lack of an efficient, general purpose...
64 CitationsSource
Cited By57
Newest
#1Majid Ghasemi (IAU: Islamic Azad University)
#2Manoochehr Kelarestaghi (Kharazmi University)H-Index: 3
Last. Arash Sharifi (IAU: Islamic Azad University)H-Index: 7
view all 4 authors...
Abstract In this paper, a robust sparse representation for medical image classification is proposed based on the adaptive type-2 fuzzy learning (T2-FDL) system. In the proposed method, sparse coding and dictionary learning processes are executed iteratively until a near-optimal dictionary is obtained. The sparse coding step aiming at finding a combination of dictionary atoms to represent the input data efficiently, and the dictionary learning step rigorously adjusts a minimum set of dictionary i...
Source
#1Abhishake (IITD: Indian Institute of Technology Delhi)H-Index: 1
#1Abhishake (IITD: Indian Institute of Technology Delhi)
Last. S. Sivananthan (IITD: Indian Institute of Technology Delhi)
view all 2 authors...
Abstract In this paper, we study the Nystrom type subsampling for large-scale kernel methods to reduce the computational complexities of big data. We discuss the multi-penalty regularization scheme based on Nystrom type subsampling which is motivated from well-studied manifold regularization schemes. We develop a theoretical analysis of the multi-penalty least-square regularization scheme under the general source condition in vector-valued function setting, therefore the results can also be appl...
1 CitationsSource
#1Quan Zhou (NUPT: Nanjing University of Posts and Telecommunications)H-Index: 9
#2Jie Cheng (Huawei)H-Index: 9
Last. Longin Jan Latecki (TU: Temple University)H-Index: 43
view all 9 authors...
Visual saliency detection plays a significant role in the fields of computer vision. In this paper, we introduce a novel saliency detection method based on weighted linear multiple kernel learning (WLMKL) framework, which is able to adaptively combine different contrast measurements in a supervised manner. As most influential factor is contrast operation in bottom-up visual saliency, an average weighted corner-surround contrast (AWCSC) is first designed to measure local visual saliency. Combined...
2 CitationsSource
#1Liu Huaping (THU: Tsinghua University)H-Index: 3
#1Huaping Liu (THU: Tsinghua University)H-Index: 32
Last. Fuchun Sun (THU: Tsinghua University)H-Index: 46
view all 4 authors...
Abstract Many multi-modal data suffers from significant weak-pairing characteristics, i.e., there is no sample-to-sample correspondence between modalities, rather classes of samples in one modality correspond to classes of samples in the other modality. This provides great challenges for the cross-modal learning for retrieval. In this work, our focus is learning cross-modal representations with minimal class label supervision and without correspondences between samples. To tackle this challengin...
2 CitationsSource
#1Fatemeh Alavi (Shiraz University)H-Index: 1
#2Sattar Hashemi (Shiraz University)H-Index: 18
Kernel learning is a challenging issue which has been vastly investigated over the last decades. The performance of kernel-based methods broadly relies on selecting an appropriate kernel. In machine learning community, a fundamental problem is how to model a suitable kernel. The traditional kernels, e.g., Gaussian kernel and polynomial kernel, are not adequately flexible to employ the information of the given data. Classical kernels are unable to sufficiently depict the characteristics of data s...
Source
#1Fatemeh Taibi (Shahid Chamran University of Ahvaz)H-Index: 1
#2Gholamreza Akbarizadeh (Shahid Chamran University of Ahvaz)H-Index: 15
Last. Ebrahim Farshidi (Shahid Chamran University of Ahvaz)H-Index: 10
view all 3 authors...
In this paper, the main goal is to identify the sine fractures of reservoir rock automatically. Therefore, a five-step algorithm is applied on the imaging logs. The first step consists of extracting the features of the imaging log by applying the Zernike moments. In the second step, the features are learned by using sparse coding. In the third step, the imaging log is segmented by using the self-organizing map neural network and the training dataset. In the fourth step, the fracture points are e...
10 CitationsSource
#1Li ShangH-Index: 6
#2Yan ZhouH-Index: 2
Last. Zhan-Li Sun (Anda: Anhui University)H-Index: 6
view all 3 authors...
To extract the essential features from a relatively small number of sampling set and further improve the feature recognition precision of images, a novel palm recognition method using the adaptive lifting wavelet transform (ALWT) based sparse representation (SR) algorithm is proposed here. This lifting wavelet behaves local texture features in spatial and the fast operation speed. While SR method can effectively represent structure features of images and behaves adaptive denoising characteristic...
Source
#1Euijoon Ahn (USYD: University of Sydney)H-Index: 9
#2Ashnil Kumar (USYD: University of Sydney)H-Index: 15
Last. Jinman Kim (USYD: University of Sydney)H-Index: 21
view all 5 authors...
Abstract The availability of large-scale annotated image datasets and recent advances in supervised deep learning methods enable the end-to-end derivation of representative image features that can impact a variety of image analysis problems. Such supervised approaches, however, are difficult to implement in the medical domain where large volumes of labelled data are difficult to obtain due to the complexity of manual annotation and inter- and intra-observer variability in label assignment. We pr...
5 CitationsSource
#1Jianqiang Song (Xidian University)H-Index: 1
#1Jianqiang Song (Xidian University)H-Index: 1
Last. Weisheng Dong (Xidian University)H-Index: 23
view all 4 authors...
Abstract Discriminative dictionary learning (DDL) has demonstrated significantly improved performance for image classification. However, most of the existing DDL methods just adopt the single-layer dictionary learning architecture, which narrows the discriminative ability of the coding vectors. Another limitation of these methods is that the atoms of the learned dictionary are easily affected by the noise in the original data. To this end, a powerful architecture, called the multi-layer discrimi...
4 CitationsSource
Jul 1, 2019 in IJCNN (International Joint Conference on Neural Network)
#1Ali Pezeshki (CSU: Colorado State University)H-Index: 16
#2Mahmood R. Azimi-Sadjadi (CSU: Colorado State University)H-Index: 26
Last. Christopher Robbiano (CSU: Colorado State University)H-Index: 1
view all 3 authors...
In this paper, a new multiple kernel machine with in-situ learning capability is introduced for pattern classification using sparse representation framework. A vector-valued score function for a given pattern is generated as a linear combination of multiple kernel functions, each of which measures the similarity between the pattern to be classified and the representative samples from every class. The representative samples for different classes are selected through the sparse approximations, by ...
Source