High Resolution Local Structure Constrained Image Upsampling Yang Zhao, Ronggang Wang*, Member, IEEE, Wenmin Wang, and Wen Gao, Fellow, IEEE School of Electronic and Computer Engineering, Peking University Shenzhen Graduate School, 2199 Lishui Road, Shenzhen 518055, China e-mail: [email protected]; [email protected]; [email protected]; [email protected] Abstract: With the development of ultra-high resolution display devices, fine texture details are becoming more and more important for visual perception. High quality image upsampling with low cost is still a challenging work. Recent upsampling methods focused on edge enhancement or try to recover high frequency components with example-based method. They either cannot recover fine texture details or are very time consuming. In this paper, we propose a fast and efficient image upsampling method by utilizing high resolution (HR) local structure constraints. We utilize average local difference to segment the low resolution (LR) image into sharp edge area and texture area, and these two areas are upsampled respectively with different HR constraints. For the sharp edge area upsampling, the HR gradient map is estimated as extra constraint to recover sharp and natural edges; For the texture area upsampling, the HR local texture structure is estimated as extra constraint to reconstruct the fine texture details. Finally, the upsampled sharp edge area and texture area are combined to get the HR image. Experimental results demonstrate the effectiveness of our method over state-of-the-art methods. Experimental Results: Video Demo:
|
AuthorLove Computer Vision Archives
December 2018
Categories |