View : 849 Download: 0

Full metadata record

DC Field Value Language
dc.contributor.advisor윤정호-
dc.contributor.author최예지-
dc.creator최예지-
dc.date.accessioned2016-08-26T04:08:56Z-
dc.date.available2016-08-26T04:08:56Z-
dc.date.issued2016-
dc.identifier.otherOAK-000000121587-
dc.identifier.urihttps://dspace.ewha.ac.kr/handle/2015.oak/214090-
dc.identifier.urihttp://dcollection.ewha.ac.kr/jsp/common/DcLoOrgPer.jsp?sItemId=000000121587-
dc.description.abstractThis paper proposes a non-linear image up-sampling method in the presence of impulsive (e.g., salt-and-pepper) noises which behave as outliers in the given data. The moving least squares (MLS) method is a very useful tool for image interpolation and superresolution. However, it is well-known in the literature that a method based on l_(2)-norm minimization shows weakness against outliers. Whereas a method based on l_(1) minimization shows better performance in image denoising. In this regards, this paper introduces a H_(d) measurement which mimics the l_(1)-measurement and then develop a moving least H_(d) (MLH) method for image interpolation. Since a method based on l_(1)-norm (e.g., TV-minimization) eliminates outliers effectively, the MLH method also has some inherent outlier removing property. On the other hand, a conventional linear method does not reflect the characteristic of a given data such that the local features of the given image are smoothed or blurred throughout the zooming process. For this reason, we employ data-adapted weight functions that consider edge orientations in the image. Accordingly, the proposed method assigns higher weights to data values which lie along the edge direction around the evaluation point. Therefore, the data-adapted MLH method simultaneously sharpens and denoises an image through the interpolating process. Experimental results are presented and compared visually and numerically by using two quantitative fidelity measurements: the PSNR(Peak Signal-to-Noise-Ratio) and the MSSIM(Mean of Structural SIMilarity). The results demonstrate the advantages of the proposed method in image up-sampling and denoising aspect.;이 논문은 노이즈가 있는 이미지를 효율적으로 확대하는 비선형 방법을 소개한다. 소금-후추 노이즈는 디지털 이미지의 획득, 전송과정에서 생기는 노이즈 중의 하나로 이는 주어진 데이터 중 아웃라이어 역할을 한다. 따라서 이미지를 확대할 때 아웃라이어에 민감하지 않은 MLH 방법을 사용하여 데이터를 부드럽게 근사하도록 한다. 하지만 기존 MLH 방법은 계산하고자 하는 점과의 거리에 따라 가중치가 정해지는 가중치함수를 사용하기 때문에, 데이터의 특성을 반영하지 못한다. 이를 보완하기 위하여, 우리가 제안하는 방법은 주어진 데이터의 지역적 특성에 따라 가중치를 갖는 데이터 반영 가중치 함수를 적용한다. 결과적으로 edge 보존효과와 디노이징 효과를 동시에 만족시키며 이미지를 효과적으로 확대할 수 있다. 마지막으로 여러 가지 테스트 이미지에 적용한 시각적인 결과와 수치적인 결과를 비교하여 이 논문에서 제안하는 방법의 이점을 보여준다.-
dc.description.tableofcontents1 Introduction 1 2 Preliminaries 5 2.1 Moving Least Squares approximation (MLS) 5 2.2 Moving Least Hd approximation (MLH) 7 3 Data-adapted Moving Least Hd 11 3.1 The convergence and the smoothness of MLH method 11 3.2 Approximation of gradients by MLS 12 3.3 A data-adapted weight function 15 4 Experimental Results 19 5 Conclusion 30 References 31 국문초록 34-
dc.formatapplication/pdf-
dc.format.extent5752390 bytes-
dc.languageeng-
dc.publisher이화여자대학교 대학원-
dc.subject.ddc500-
dc.titleImage Zooming Method based on Data-adapted Moving Least Hd for Salt-and-Pepper Noise-
dc.typeMaster's Thesis-
dc.format.pageii, 34 p.-
dc.contributor.examiner민조홍-
dc.contributor.examiner이준엽-
dc.contributor.examiner윤정호-
dc.identifier.thesisdegreeMaster-
dc.identifier.major대학원 수학과-
dc.date.awarded2016. 2-
Appears in Collections:
일반대학원 > 수학과 > Theses_Master
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

BROWSE