Cited time in webofscience Cited time in scopus

Enhanced Image Stitching using Recurrent Elastic Warps and High-Definition Neural Imaging

Title
Enhanced Image Stitching using Recurrent Elastic Warps and High-Definition Neural Imaging
Alternative Title
신경망을 통한 고화질 이미징과 순환적 탄성 워핑을 이용한 강화된 이미지 스티칭
Author(s)
Minsu Kim
DGIST Authors
Minsu KimKyong Hwan JinSunghoon Im
Advisor
진경환
Co-Advisor(s)
Sunghoon Im
Issued Date
2023
Awarded Date
2023-08-01
Type
Thesis
Description
Image Stitching; Implicit Neural Representation; Elastic Warp
Table Of Contents
I. INTRODUCTION 1
1.1 Introductory Remarks 1
1.2 Failure-tolerant Elastic Image Alignment 1
1.3 High-Definition Image Stitching 4
II. ELASTIC IMAGE ALIGNMENT 6
2.1 Problem Formulation 6
2.2 Neural Networks for Elastic warps Estimation 7
2.2.1 Overview of Cells 7
2.2.2 State Initialization 7
2.2.3 Deformation Field Evaluation 9
2.3 Iterative Homography Prediction 10
2.4 Iterative Thin-plate Spline Prediction 10
2.5 Loss Functions and Training Strategy 11
2.5.1 Global Alignment Network 11
2.5.2 Local Alignment Network 11
2.6 Implementation Details 11
2.6.1 Encoders 11
2.6.2 Adaptor CNNs 12
2.6.3 Regressors in Cells 12
2.7 Image Stitching 12
III. HIGH-DEFINITION IMAGE STITCHING 13
3.1 Problem Formulation 13
3.2 Continuous Neural Filter for Detail-aware Feature Warps 13
3.2.1 Neural Warping 13
3.2.2 Merging Fourier Features 14
3.3 Training Details 15
3.3.1 Scratch Training 15
3.3.2 Fine-tuning 15
3.4 Architecture Details 16
3.4.1 Neural Image Stitching 16
IV. APPLYING ON SYNTHETIC DATASET 17
4.1 Efficiency of Recurrent Neural Network 17
4.1.1 Data Preparation 17
4.1.2 Experimental Setup 17
4.1.3 Quantitative Comparison 18
4.1.4 Evaluation on Low-light Images 18
4.1.5 Conclusion 19
4.2 High-Definition Image Stitching 20
4.2.1 Data Preparation 20
4.2.2 Experimental Setup 20
4.2.3 Quantitative Comparison 21
4.2.4 Qualitative Comparison 21
4.2.5 Fourier-based latent space 22
4.2.6 Ablation Study 22
4.2.7 Conclusion 23
V. APPLYING ON REAL DATASET 24
5.1 Elastic Image Alignment 24
5.1.1 Data Preparation 24
5.1.2 Experimental Setup 24
5.1.3 Quantitative Comparison 25
5.1.4 Qualitative Comparison 28
5.1.5 Ablation Study 28
5.1.6 Discussion 29
5.1.7 Conclusion 29
5.2 High-Definition Image Stitching 30
5.2.1 Data Preparation 30
5.2.2 Experimental Setup 30
5.2.3 Quantitative Comparison 30
5.2.4 Qualitative Comparison 33
5.2.5 Model Specifications 33
5.2.6 Ablation Study 34
5.2.7 Validity of Neural Warping 34
5.2.8 Conclusion 35
References 36
국문요약 39
URI
http://hdl.handle.net/20.500.11750/46446

http://dgist.dcollection.net/common/orgView/200000684911
DOI
10.22677/THESIS.200000684911
Degree
Master
Department
Department of Electrical Engineering and Computer Science
Publisher
DGIST
Related Researcher
  • 임성훈 Im, Sunghoon
  • Research Interests Computer Vision; Deep Learning; Robot Vision
Files in This Item:

There are no files associated with this item.

Appears in Collections:
Department of Electrical Engineering and Computer Science Theses Master

qrcode

  • twitter
  • facebook
  • mendeley

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

BROWSE