Publications

Detailed Information

Robust Depth Map Estimation for Radiometrically Varying Stereo Images : 스테레오 이미지의 조명 변화에 강인한 변위 지도 예측 방법

DC Field Value Language
dc.contributor.advisor이상욱-
dc.contributor.author허용석-
dc.date.accessioned2017-07-13T06:55:25Z-
dc.date.available2017-07-13T06:55:25Z-
dc.date.issued2012-08-
dc.identifier.other000000004862-
dc.identifier.urihttps://hdl.handle.net/10371/118870-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 전기·컴퓨터공학부, 2012. 8. 이상욱.-
dc.description.abstract스테레오 정합의 목표는 시점 또는 시간차를 두고 얻은 이미지들 사이에서 정합 점을 예측함으로써, 깊이 정보 또는 3차원 정보를 획득하는 것을 목표로 하고 있다. 일반적으로 대부분의 스테레오 정합 알고리즘들은 이러한 입력 이미지들이 라디오메트릭하게 (radiometrically) 조정이 되어 있다고 가정하고 있다. 이러한 이미지들에 대해서는, 밝기 값의 절대 차이 값 (absolute difference)과 같은 간단한 정합 비용으로도 스테레오의 성능이 크게 떨어지지 않는다. 그러나, 실제 여러 상황의 이미지에서는 여러 라디오메트릭 변화 (radiometric variation)가 일어나기 때문에, 대부분의 스테레오 알고리즘의 성능은 이러한 실제 상황에서는 크게 성능 저하가 일어나게 된다.

본 학위 논문에서는, 이러한 문제를 해결하기 위한 강인한 깊이 정보 예측 알고리즘들을 제안한다. 첫번째로, 이러한 라디오메트릭 변화에 무관한 새로운 스테레오 정합 비용 (measure)을 제안하였다. 이를 위해서, 컬러의 형성 모델을 보다 구체적으로 살펴보았고, Adaptive Normalized Cross Correlation (ANCC)라고 하는 새로운 비용 (measure)을 제안하였다. 두번째로, SIFT 묘사기 (descriptor)와 상호 정보량 (mutual information)을 이용하는 새로운 강인한 깊이 정보 예측 알고리즘을 제안하였다. 이를 위해서 SIFT 묘사기와 상호 정보량의 상호 보완성의 성질을 이용하였다. 세번째로, 입력 스테레오 이미지의 컬러를 일치시키는 것과 깊이 정보를 예측하는 것을 반복적으로 수행하는 새로운 방법을 제안하였다. 마지막으로, 이미지에 매우 어두운 부분이나 매우 밝은 부분이 존재할 때 발생하는 정합 문제를 해결하기 위하여, high dynamic range imaging (HDRI)과 스테레오 정합을 동시에 수행하는 방법을 제안하였다. 제안하는 방법들이 기존의 방법보다 우수함을 여러 가지 실험을 통해서 정량적, 정성적으로 분석하였다.
-
dc.description.abstractStereo matching aims to obtain depth map or 3D information by finding correct
correspondence between images captured from different point of views or at different times.
In general, most stereo algorithms assume radiometrically calibrated images. For these calibrated images, simple matching costs such as absolute difference of intensities do not degrade the performance of stereo algorithms. However, there exist many real and practical situations or challenging applications in which radiometric variations between stereo images are inevitable.
The performance of most conventional stereo matching algorithms can be severely
degraded under these radiometric variations.

In this thesis, we propose several robust depth map estimation and color correction methods for solving these problems. Firstly, we present a new stereo matching measure that is insensitive to radiometric variations between left and right images. For this goal, we use the color formation model explicitly in our
framework and propose a new measure called the Adaptive Normalized
Cross Correlation (ANCC) for a robust and accurate correspondence
measure. Secondly, we propose a more robust method based on mutual information (MI) combined with SIFT descriptor to find correspondence for images which undergo various radiometric changes. Although the proposed ANCC is robust to radiometric variations to some degree, it still has some limitations. To devise a more robust measure, we combined MI and SIFT that have complementary roles to each other. Thirdly, we propose a new method which infers both accurate depth maps and color-consistent stereo images for radiometrically varying stereo images. From our framework, we showed that both depth map estimation and color correction could be beneficial to each other. Finally, we also propose an efficient method to generate high dynamic range multi-view stereo images as well as corresponding accurate depth maps for alternative exposure multi-view stereo images that consist of consecutive long and short exposure stereo images. Experimental results show that our methods outperform other state-of-the-art stereo methods under severely different radiometric conditions between stereo images.
-
dc.description.tableofcontents1 Introduction 1
1.1 Motivation 1
1.2 Related works 4
1.3 Outline of the Thesis 6
2 Robust Stereo Matching Using Adaptive Normalized Cross Correlation 11
2.1 Introduction 11
2.2 Stereo Energy Formulation 12
2.3 Color Invariant Information 13
2.3.1 Color image formation model 15
2.3.2 Color image normalization 16
2.4 Stereo Matching using ANCC 17
2.4.1 Overview of our approach 19
2.4.2 Log-chromaticity normalization 20
2.4.3 Adaptive normalized cross correlation 22
2.4.4 Combining log-chromaticity and original RGB colors 24
2.4.5 Global energy modeling 25
2.5 Experimental Results 26
2.5.1 Light source changes 27
2.5.2 Camera exposure changes 31
2.5.3 Gamma correction changes 35
2.5.4 Noise variations 37
2.5.5 Application to aerial image 39
2.6 Discussion 42
2.6.1 Effects of adaptive-weight in ANCC 42
2.6.2 Effect of only similarity measure 46
2.6.3 Performance assessment based on parameter values and comparison of computational time 47
2.7 Conclusions 48
3 Mutual Information-based Stereo Matching Combined with SIFT Descriptor in Log-chromaticity Color Space 51
3.1 Introduction 51
3.2 Mutual Information as a Stereo Correspondence Measure in MAPMRF Framework 54
3.3 Proposed Algorithm 58
3.3.1 Transformation to log-chromaticity color space and discretization 59
3.3.2 Joint probability using SIFT descriptor 60
3.3.3 Disparity map estimation in MAP-MRF62
3.4 Experimental Results 63
3.4.1 MI vs. SIFT 65
3.4.2 Different exposures 65
3.4.3 Different configurations of the light source 66
3.5 Conclusion 68
4 Joint Depth Map and Color Consistency Estimation for Radiometrically Varying Stereo Images 75
4.1 Introduction 75
4.2 Proposed Algorithm 77
4.2.1 Transformation to log-chromaticity color space and quantization 79
4.2.2 Joint probability density function and linear function estimation in log-chromaticity color 80
4.2.3 Disparity map estimation 82
4.2.4 Occlusion map estimation 88
4.2.5 Color consistency via stereo color histogram equalization 89
4.2.6 Boosting the disparity map estimation using SCHE images 91
4.3 Experimental Results 92
4.3.1 Performance of iterative joint boosting of disparity map and color consistency according to the iteration 93
4.3.2 Color consistency performance 95
4.3.3 Analysis on stereo matching 98
4.3.4 Stereo matching performance comparison 99
4.3.5 Tests for scenes with different cameras 103
4.3.6 Tests for aerial images 106
4.3.7 Tests for internet images 107
4.4 Conclusion 108
5 Joint Depth Map and High Dynamic Range Image Estimation From Alternating Exposure Multi-view Stereo Images 109
5.1 Introduction 109
5.2 Proposed Algorithm 112
5.2.1 Initial Depth Map Estimation Using Stereo Images With the Same Exposure Setting 113
5.2.2 Depth Map Fusion Using Fusion Move 113
5.2.3 HDRIs Generation Using Fused Depth Maps 116
5.3 Experimental Results 119
5.4 Conclusion 120
6 Conclusion 129
Bibliography 133
-
dc.formatapplication/pdf-
dc.format.extent13845052 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subject스테레오 정합-
dc.subject조명 변화-
dc.subject카메라 변화-
dc.subject상호 정보량-
dc.subject컬러 일치성-
dc.titleRobust Depth Map Estimation for Radiometrically Varying Stereo Images-
dc.title.alternative스테레오 이미지의 조명 변화에 강인한 변위 지도 예측 방법-
dc.typeThesis-
dc.contributor.AlternativeAuthorYong Seok Heo-
dc.description.degreeDoctor-
dc.citation.pagesxxi, 146-
dc.contributor.affiliation공과대학 전기·컴퓨터공학부-
dc.date.awarded2012-08-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share