Application of Different Feature Weights Based on Learning Feature Dictionary for Image Super-Resolution
Hyun Ho Han1, Sang Hun Lee2, Jong Yong Lee3, Young Soo Park4, Ki Bong Kim5
1Hyun Ho Han, Graduate School, Kwangwoon University, Korea, East Asian.
2Sang Hun Lee, Ingenium College of liberal arts, Kwangwoon University, Korea, East Asian.
3Jong Yong Lee, Ingenium College of Liberal Arts, Kwangwoon University, Korea, East Asian.
4Young Soo Park, Ingenium College of Liberal Arts, Kwangwoon University, Korea, East Asian.
5Ki Bong Kim, Department of Computer Information, Daejeon Health Institute of Technology, Korea, East Asian.
Manuscript received on 08 June 2019 | Revised Manuscript received on 14 June 2019 | Manuscript Published on 22 June 2019 | PP: 71-76 | Volume-8 Issue-8S2 June 2019 | Retrieval Number: H10140688S219/19©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open-access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: In this paper, we propose a method of feature differential application based on dictionary data structure for the generation of a super-resolution image in a single image. The existing method of generating super-resolution based on the dictionary data structure results in poor quality, such as artifacts or the staircase. because it refers to the value of the dictionary data without analyzing the configuration of each area. In order to overcome this problem, the proposed method generates a low-resolution image for the dictionary data construction and constructs a pair of dictionary data of low resolution and high resolution through feature extraction with the original image. In order to differentially apply the dictionary features, we estimated the feature loss area in the bicubic interpolation process and analyzed the composition of the details of the area, then weighed it. Using the calculated weight values, we applied the feature data of the dictionary data to each region differentially in order to generate an improved super-resolution image. For experimentation, the original image was compared with the reconstructed image with PSNR and SSIM.
Keywords: Super Resolution, Linear Interpolation, Patch Information, Region Segmentation, PSNR.
Scope of the Article: Agent-Based Learning and Knowledge Discovery