Logo image
Self-Learning-Based Single Image Super-Resolution and Deblocking for Highly Compressed Images
Thesis

Self-Learning-Based Single Image Super-Resolution and Deblocking for Highly Compressed Images

Zhuang, Boqi
Masters, 國立清華大學, 通訊工程研究所
2012

Abstract

自我學習 超解析度 去塊效應 高壓縮影像 稀疏表示 影像重建 字典學習 圖像形態分量分析 Self Learning Super Resolution Morphological Component Analysis Dictionary Learning Sparse Representation Image Decomposition Deblocking Highly Compressed Images
Highly compressed images are usually not only of low-resolution, but also suffer from compression artifacts, e.g., blocking artifacts; ringing artifact. So if we do image super-resolution (SR) to a highly compressed image directly, we will also simultaneously magnify the artifacts, and get unpleasing visual quality. But we find that if we individually performing deblocking followed by SR to an image, it would lose some image details which may be useful for SR when deblocking, and resulting in worse SR result. If we performing SR followed by deblocking, it will magnify the blocking artifacts, and we will hardly to deblocking well for the SR result, so the result will remain blocking artifact or over smooth. In summary of above, we find that we can't get a good result if we individually performing deblocking and SR, so we want to find a method which can combine SR and deblocking in one operation. In this thesis, we propose a self-learning-based SR framework to simultaneously achieve single-image SR and blocking artifact removal for highly compressed images. In our method, we propose to self-learn image sparse representation for modeling the relationship between low and high-resolution image patches in terms of the learned dictionaries, respectively, for image patches with and without blocking artifacts. As a result, image SR and deblocking can be simultaneously achieved via sparse representation and MCA (morphological component analysis)-based image decomposition. Experimental results demonstrate the efficacy of the proposed algorithm.

Metrics

1 Record Views

Details

Logo image