CHOSUN

A Comparative Analysis of Image Denoising Techniques for Efficient Interpolation in Noisy Color Channel

Metadata Downloads
Author(s)
Ramesh Kumar Lama
Issued Date
2016
Abstract
Most of the existing image interpolation schemes assume that the low resolution image is noise free. This assumption is invalid in practice
because image is corrupted by noise during the image acquisition process.
This thesis presents a implementation of denoising and interpolation sequentially. For each noisy sample, we propose an efficient noise reduction technique designed to work with Color Filtering Array (CFA) data acquired by CCD/CMOS image sensors. CFAimage neighborhood
vectors are first projected on the principal component analysis (PCA) bases. The PCA bases are obtained by eigenvalue decomposition. The denoising is performed by the shrinkage of coefficients of patches which we obtain from principal components. The effectiveness of the shrinkage depends on the selection of threshold and transform to sparsely represent the true image data, thus separating it from the noise. The proposed method estimates the optimized threshold value based on the local
statistical characteristics of the patch. Final color image is generated by color demosaicking (CDM). Once the CDM is accomplished, the low resolution image is interpolated to fit on the high resolution display devices. In this dissertation we implement various wavelet based
interpolation schemes which are adaptive directional wavelet transform,
hybrid implementation of discrete wavelet transform and discrete cosine transform (DCT) and dual-tree complex wavelet transform (DT-CWT) and hidden Markov model (HMM). The adaptive directional wavelet transform
is the lifting implementation of wavelet transform. This method is able to transform an image along multiple directions as well as conventional horizontal and vertical orientations. In the proposed method the high-frequency sub band images are obtained by adaptive wavelet
transforms (ADWT). We interpolate high-frequency sub band images and the input low resolution image, and finally the high resolution image is generated by combining the interpolated subimages in the inverse ADWT
process. In hybrid implementation of wavelet transform and DCT, the high frequency wavelet coefficients are interpolated using the zero pad method in DCT domain. The upscaled image is generated using inverse DWT of
interpolated coefficients and original image. Finally in DT-CWT based method, we use DT-CWT is to decompose the low-resolution image into different subbands. The proposed method estimates the higher band coefficients by learning the correlation between the coefficients across the scale. In this paper, the relationship between the wavelet coefficients across the scale is described by HMM, and each wavelet coefficient is modelled
by a Gaussian mixture having multiple means and variances. Experimental results demonstrate that the proposed methods are competitive with state-of-the-art methods despite its simplicity. The proposed method is also suitable for implementation in low power mobile devices with imaging capabilities such as camera phones.|기존의 대부분의 영상 보간 방식은 노이즈가 없는 것으로 가정한다.
노이즈가 영상 복호화 과정에서 사라지기 때문에 이러한 가정은 실제로 유효하다. 이 논문은 영상의 순차적인 잡음 제거와 보간 구현 방법에 대해 설명한다. 또한, 잡음 샘플(noisy sample)을 위해, CCD/CMOS 영상 센서에서 얻어진 컬러 필터링 어레이(CFA)에 적용할 수 있는 효율적인 잡음 감소 기법을 제안한다. 화상 이웃 벡터는 PCA 베이스 상에 투사된다. PCA 베이스는 고유 분해에 의해 얻을 수 있다. 잡음 제거는 영상의 주 성분으로부터 여러번 패치된 된 계수의 수축에 의해 수행된다. 수축 효과는 임계치의 선택과 노이즈로부터 원본 영상을 분리하려는 변환 과정에 의존한다.
제안하는 방법은 패치의 통계적인 특성들에 기초하여 최적화 된 임계 값을 추정한다. 한번 CDM이 수행되고 나면, 저해상도 영상은 듀얼-트리 콤플렉스 웨이블릿 변환과 (DT-CWT) 히든 마르코프 모델(HMM)을 기초하여 고해상도의 표시 장치에 적합하도록 보간된다. 제안한 방법의 DT-CWT는 상이한 서브밴드의 저해상도 화상을 분해하는데에도 사용된다. 웨이블릿 도메인 보간에서 주어진 영상은 고해상도 영상의 웨이블릿 계수의 저주파 LL 서브밴드로 간주된다. 제안한 방법을 통해 전체 스케일 계수간 상관 관계를 학습하여 보다 큰 크기를 가진 밴드 계수들을 추정한다. 이 과정에서 스케일 사이간 웨이
블릿 계수의 관계는 HMM에 의해 기술되고, 각각의 웨이블릿 계수들은 여러 수단 및 분산을 갖는 가우시안 혼합에 의해 모델링된다.
실험 결과를 통해 제안한 알고리즘은 기존의 다른 방법들에 비해 선명한 이미지를 얻을 수 있음을 증명한다. 그리고 기존 방법보다 단순하면서 보다 향상된 성능을 보여준다.
또한 제안한 방법은 카메라 폰 등의 촬상 기능을 가진 저전력 모바일 장치에서 구현하기에 적합하다.
Alternative Title
노이즈 색상 채널에서 효율적인 보간법을 위한 영상 디노이징 기술의 비교 분석
Alternative Author(s)
라마 라메쉬 쿠마
Affiliation
Department of Information and Communication Engineering
Department
일반대학원 정보통신공학과
Advisor
Goo-Rak Kwon
Awarded Date
2016-02
Table Of Contents
TABLE OF CONTENTS i
LIST OF FIGURES iv
LIST OF TABLES viii
ACRONYMS ix
ABSTRACT (ENGLISH) x
ABSTRACT (KOREAN) xii
1. Introduction 1
1.1. Contribution of Dissertation 3
1.2. Organization of Dissertation 4
2. Image acquisition 5
2.1. Introduction to CFA Images 5
2.2. Raw Image Noise Model 7
2.3. Noise on CFA Sensor 9
3. System to Reduce CFA Noise 13
3.1. Noise on CFA Sensor 13
3.2. Literature review on CFA image denoising 13
3.3. Principal Component Models 16
3.3.1 Singular Value Decomposition (SVD) 18
3.3.2 Mathematical Definition of SVD 18
3.3.3 Relationship bewtween PCA and SVD 19
3.4. System to Reduce CFA Noise 19
3.4.1 Patch based Denoising in Principal Component
Analysis Domain 20
4. Performance Evaluation 26
4.1 CFA Image Denoising 27
4.1.1 Color Interpolation and CFA Image Denoising 28
4.2 Subsection Summary 34
5. Image Interpolation 35
5.1 Introduction 35
5.2 Literature Review on Image Interpolation 36
5.3 Isotropic 36
5.3.1. Nearest Neighbor 36
5.3.2. Bilinear 37
5..3.3 Bicubic 38
5.4 Edge directed interpolation 39
5.4.1. Modified edge directed interpolation 42
5.5 Transform based interpolation 42
6. Wavelet Transform 43
6.1 The wavelet transform 43
6.2 Wavelet Characteristics 43
6.3 Wavelet Analysis 44
6.4 Wavelet Transform 45
6.4.1 Discrete Wavelet Transform (DWT) 46
6.5 2D Wavelet Transform 48
6.6 Wavelet Transform based Interpolation 50
6.6.1 General Approach 50
6.7 Lifting based Wavelet Transform 52
6.7.1 Adaptive Directional Lifting 55
6.7.2 ADL based Interpolation 57
6.7.3 Discrete Cosine Transform 57
6.7.4 One Dimensional DCT 57
6.8 Discrete Cosine Transform based Image Resizing 58
6.9 Interpolation Using the Combination of DWT and DCT 60
6.10 Performance Evaluation 64
6.11 Subsection Summary 72
7 Complex Wavelet Transform based Interpolation 73
7.1 Complex Wavelets 73
7.2 The Dual Tree Complex Wavelet Transform 73
7.3 Statistical Modeling of Complex Wavelt Transform 77
7.4 Performance Evaluation 81
7.5 Subsection Summaries 91

8. Conclusion 92

Bibliography 93

Acknowledgement 100
Degree
Doctor
Publisher
Chosun University, Department of Information and Communication Engineering
Citation
Ramesh Kumar Lama. (2016). A Comparative Analysis of Image Denoising Techniques for Efficient Interpolation in Noisy Color Channel.
Type
Dissertation
URI
https://oak.chosun.ac.kr/handle/2020.oak/12584
http://chosun.dcollection.net/common/orgView/200000265163
Appears in Collections:
General Graduate School > 4. Theses(Ph.D)
Authorize & License
  • AuthorizeOpen
  • Embargo2016-02-25
Files in This Item:

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.