Quality-efficient Wiener Filter- and Context-consistency-based Image Interpolation

碩士 === 國立臺灣科技大學 === 資訊工程系 === 103 === With the rapid development of consumer electronics and multimedia technology, high-resolution display, tablet PC and smart phone are becoming the current main-stream consumer electronic products.In order to show the image of low resolution on a high-resolution d...

Full description

Bibliographic Details
Main Authors: JIE ZHENG, 鄭傑
Other Authors: Kuo-Liang Chung
Format: Others
Language:zh-TW
Published: 2015
Online Access:http://ndltd.ncl.edu.tw/handle/39359179266832338022
Description
Summary:碩士 === 國立臺灣科技大學 === 資訊工程系 === 103 === With the rapid development of consumer electronics and multimedia technology, high-resolution display, tablet PC and smart phone are becoming the current main-stream consumer electronic products.In order to show the image of low resolution on a high-resolution device clearly, developing an effective image zoom technique is an important research issue.Based on the improved Wiener filter, this thesis presents an effective image interpolation technique to enhance the quality of the enlarged image.In the proposed image interpolation technique, we first incorporate the con-text-consistency with the conventional Wiener filter and then conduct it by an iterative process so as to improve the accuracy of the least square method. In addition, this thesis also presents two prediction modes which are applicable to different textures, respectively.In the experiments, for the image with simple horizontal and vertical textures, we find that the traditional bicubic interpolation technique can deliver better enlarged quality. In order to fuse the advantages of both the two proposed prediction modes and the bicubic interpolation, we develop an effective evaluation function to determine the best fusion weights; after that, quality-superior high-resolution images can be constructed by the proposed image zoom technique. Experimental results show that the proposed image zoom technique does deliver better enlarged quality in terms of PSNR, SSIM, and visual effect when compared with the three well-known interpolation techniques, NEDI, SAI, and RSAI.