Immersive Interactive SAR Image Representation Using Non-negative Matrix Factorization
TL;DR: This work utilizes a 3-D interactive application to support user-images interactions to generate compact and informative features from images content and proposes a divide-and-conquer approach to cluster a massive amount of images using a small subset of interactions.
read more
Abstract: Earth observation (EO) images clustering is a challenging problem in data mining, where each image is represented by a high-dimensional feature vector. However, the feature vectors might not be appropriate to express the semantic content of images, which eventually lead to poor results in clustering and classification. To tackle this problem, we propose an interactive approach to generate compact and informative features from images content. To this end, we utilize a 3-D interactive application to support user-images interactions. These interactions are used in the context of two novel nonnegative matrix factorization (NMF) algorithms to generate new features. We assess the quality of new features by applying k-means clustering on the generated features and compare the obtained clustering results with those achieved by original features. We perform experiments on a synthetic aperture radar (SAR) image dataset represented by different state-of-the-art features and demonstrate the effectiveness of the proposed method. Moreover, we propose a divide-and-conquer approach to cluster a massive amount of images using a small subset of interactions.
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Figures
Fig. 1. (a) the visualization of the clustering result in the CAVE. Images are positioned around their cluster centers based on their distances. A sample image of each cluster is used to depict the cluster center. (b), (c) show user interactions on a desktop. A mis-clustered image is connected to a semantically cluster center by a green line. (d) a mis-clustered image (the image with red border) is connected (green line) to the cluster center of target cluster (with blue border). This interaction updated the semantic similarity matrix W , which is used in our novel NMF algorithms. 
Fig. 5. Clustering results of the proposed divide-and-conquer approach as a function of the number of interactions, represented by accuracy (first row) and normalized mutual information (second row). The first, second and third columns show the results SAR images represented by Mean-Variance, Image intensity and WLD features, respectively. 
Fig. 6. Clustering results of different subspaces (new features) represented by accuracy (first row) and normalized mutual information (second row). The first, second and third columns show the results of SAR images represented by Mean-Variance, Image intensity and WLD features, respectively. 
Fig. 8. The convergence speed for NMF, VNMF and CMNMF applied on three features; a)Mean-Variance; b) Image Intensity; c) WLD. 
Fig. 7. Exemplary images of four randomly chosen classes after applying k-means clustering to original features (left column) and the new features obtained by VNMF (middle column) and CMNMF (right column). Each row shows sample images of a random chosen class. Green borders depict correct clustered images and the red ones depict incorrect clustered images. 
Fig. 2. Sample images from the SAR data set. There are 15 images, each one is representing one class.
Citations
•Proceedings Article
Proceedings of the twenty-first international conference on Machine learning
Carla E. Brodley
- 04 Jul 2004
TL;DR: It is shown that backfitting --- a traditional non-parametric, yet highly efficient regression tool --- can be derived in a novel formulation within an expectation maximization (EM) framework and thus can finally be given a probabilistic interpretation.
1.1K
Sparse Synthetic Aperture Radar Imaging From Compressed Sensing and Machine Learning: Theories, applications, and trends
TL;DR: In this paper , the sparse SAR image formation can be treated as a class of ill-posed linear inverse problems, and the resolution is limited by the data bandwidth for traditional imaging techniques via matched filter (MF).
106
Sparse Synthetic Aperture Radar Imaging From Compressed Sensing and Machine Learning: Theories, applications, and trends
TL;DR: In this paper , the sparse SAR image formation can be treated as a class of ill-posed linear inverse problems, and the resolution is limited by the data bandwidth for traditional imaging techniques via matched filter (MF).
78
SAR Target Recognition via Local Sparse Representation of Multi-Manifold Regularized Low-Rank Approximation
TL;DR: A low-dimensional representation model via incorporating multi-manifold regularization term into the low-rank matrix factorization framework and local sparse representation is proposed for classification to improve the discriminative ability of target recognition under EOCs.
60
GEO Targets ISAR Imaging With Joint Intra-Pulse and Inter-Pulse High-Order Motion Compensation and Sub-Aperture Image Fusion at ULCPI
Sihua Shao,Hongwei Liu,Jiaqi Wei +2 more
TL;DR: High-resolution ISAR imaging of GEO targets at ULCPI with joint intra-pulse and inter-pulse high-order motion compensation and sub-aperture image fusion
5
References
Nonlinear dimensionality reduction by locally linear embedding.
Sam T. Roweis,Lawrence K. Saul +1 more
TL;DR: Locally linear embedding (LLE) is introduced, an unsupervised learning algorithm that computes low-dimensional, neighborhood-preserving embeddings of high-dimensional inputs that learns the global structure of nonlinear manifolds.
A global geometric framework for nonlinear dimensionality reduction.
TL;DR: An approach to solving dimensionality reduction problems that uses easily measured local metric information to learn the underlying global geometry of a data set and efficiently computes a globally optimal solution, and is guaranteed to converge asymptotically to the true structure.
Learning the parts of objects by non-negative matrix factorization
TL;DR: An algorithm for non-negative matrix factorization is demonstrated that is able to learn parts of faces and semantic features of text and is in contrast to other methods that learn holistic, not parts-based, representations.
14.2K
Learning parts of objects by non-negative matrix factorization
D. D. Lee
- 01 Jan 1999
TL;DR: In this article, non-negative matrix factorization is used to learn parts of faces and semantic features of text, which is in contrast to principal components analysis and vector quantization that learn holistic, not parts-based, representations.
9.6K
•Proceedings Article
Algorithms for Non-negative Matrix Factorization
Daniel D. Lee,H. Sebastian Seung +1 more
- 01 Jan 2000
TL;DR: Two different multiplicative algorithms for non-negative matrix factorization are analyzed and one algorithm can be shown to minimize the conventional least squares error while the other minimizes the generalized Kullback-Leibler divergence.
Related Papers (5)
Deng Cai,Hujun Bao,Xiaofei He +2 more
- 20 Jun 2011
Zechao Li,Jinhui Tang +1 more
- 03 Dec 2015
Juan C. Caicedo,Fabio A. González +1 more
- 03 Sep 2012