Image super resolution using Gaussian Process Regression with patch clustering

Authors:
Fang Xie;Cheng Deng;Jie Xu;Jifei Yu;Jie Li
Affiliations:
Xidian University, Xi'an, China;Xidian University, Xi'an, China;Xidian University, Xi'an, China;Xidian University, Xi'an, China;Xidian University, Xi'an, China
Venue:
Proceedings of the Fifth International Conference on Internet Multimedia Computing and Service
Year:
2013

Citing 8
Cited 0

Example-Based Super-Resolution

IEEE Computer Graphics and Applications
Gaussian Processes for Machine Learning (Adaptive Computation and Machine Learning)

Gaussian Processes for Machine Learning (Adaptive Computation and Machine Learning)
Large Scale Multiple Kernel Learning

The Journal of Machine Learning Research
Generalizing the Nonlocal-means to super-resolution reconstruction

IEEE Transactions on Image Processing
Single-Image Super-Resolution Using Sparse Regression and Natural Image Prior

IEEE Transactions on Pattern Analysis and Machine Intelligence
Single image super-resolution using Gaussian process regression

CVPR '11 Proceedings of the 2011 IEEE Conference on Computer Vision and Pattern Recognition
New edge-directed interpolation

IEEE Transactions on Image Processing
Image Superresolution Using Support Vector Regression

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In Super-resolution (SR) community, Gaussian Process Regression (GPR) has been recognized as an effective non-parametric Bayesian approach to predict nonlinear relationship between a low-resolution (LR) image and its corresponding high-resolution (HR) estimation. However, modeling the pixel-wise relationship is expansive and is not necessary for structural redundancy in natural images. So, we propose a novel image super resolution approach based on GPR model. Specifically, we first learn GPR model between LR image and its corresponding high frequency details in the HR image. During the learning GPR model, the parameters of the covariance function in the GPR cannot accurately describe the local geometric structures especially when all patches with the same sizes in a single LR image are processed as a whole. To solve the problem, we utilize K-means clustering algorithm to group all the training pixel-patch samples according to the local geometric structure of LR patches. In each category, GPR model is learned from the training database. Experimental results demonstrate that our algorithm gains a significant improvement in terms of the quality of super-resolution.