A Super-Resolution Algorithm using Patch Structure Matching

• Main Authors: Peng-YuChen, 陳鵬宇
• Other Authors: Shen-Chuan Tai
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/76449690143335516424

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 101 === The main purpose of super resolution technology is to generate high-resolution (HR) images from low-resolution (LR) images. In this thesis, a vector quantization (VQ) based super resolution algorithm is proposed to produce HR images. Firstly, the initial blurred HR images are generated by a simple interpolation method. Furthermore, the high-frequency information images are obtained by searching the pre-trained codebook to find the best matching codevector. The final enlarged images are generated by combining the initial blurred images and the high-frequency information images. In order to predict the high-frequency information accurately, a patch structure matching method is proposed in the codebook searching phase. Besides, LBG training algorithm is also modified to adapt to the patch structure matching. Experimental results show that the proposed algorithm produces HR images with better visual quality.