Depth-Guided Full-Focus Super-Resolution Network for Light Field Images DOI
Deqian Kong, Yan Yuan, Lijuan Su

et al.

Published: Jan. 19, 2024

Light field (LF) imaging system captures the two-dimensional (2D) spatial and 2D angular information of scenes within a single exposure time. Due to this distinctive feature, technique has been rapidly developed over past two decades. However, LF images suffer from low resolution. Currently, numerous deep learning (DL)-based approaches have employed address issue. existing super-resolution (SR) networks ignore defocus blur caused by depth variations, fail yield high-resolution (HR) full-focus directly processing with information. In paper, tackle challenge, we propose new SR method reconstruct HR low-resolution (LR) multi-defocus images. To accomplish task, The degraded dataset is generated utilizing intrinsic as guidance designing spatially-variable (SV) degradation method. designed parts: depth-guided image partitioning process degradation-prior-SR network. Experimental results indicated that our outperforms other both quantitatively qualitatively.

Language: Английский

How I Met Your V2X Sensor Data: Analysis of Projection-Based Light Field Visualization for Vehicle-to-Everything Communication Protocols and Use Cases DOI Creative Commons
Peter A. Kara, András Wippelhauser, Tibor Balogh

et al.

Sensors, Journal Year: 2023, Volume and Issue: 23(3), P. 1284 - 1284

Published: Jan. 22, 2023

The practical usage of V2X communication protocols started emerging in recent years. Data built on sensor information are displayed via onboard units and smart devices. However, perceptually obtaining such data may be counterproductive terms visual attention, particularly the case safety-related applications. Using windshield as a display solve this issue, but switching between 2D 3D reality traffic introduce issues its own. To overcome difficulties, automotive light field visualization is introduced. In paper, we investigate use cases projection-based technology. Our work motivated by abundance data, low latency transfer, availability prototypes, prevalent dominance non-autonomous non-remote driving, lack V2X-based solutions. As our primary contributions, provide comprehensive technological review communication, set recommendations for design implementation, an extensive discussion implication analysis, exploration utilization based standardized protocols, use-case-specific considerations.

Language: Английский

Citations

5

Boosting Light Field Image Super Resolution Learnt From Single-Image Prior DOI
Xingzheng Wang, Zixuan Wang, Wenhao Huang

et al.

IEEE Transactions on Computational Imaging, Journal Year: 2023, Volume and Issue: 9, P. 1139 - 1151

Published: Jan. 1, 2023

In recent years, many deep learning networks are proposed for light field super resolution (LFSR). LFSR problem is essentially ill-posed since unknown detail information need to be predicted. Hence require plentiful content (e.g., shape, color, texture) learned from sufficiently diverse scenarios. However, due the high collection cost, existing datasets in small size and have few scenarios, which could not meet requirement limit performance of networks. To solve this problem, we a novel framework significantly boost their performance. Our main idea introduce valuable single images into as prior. Specifically, first, view synthesis method applied add unreal disparity images, increasing dimensionality hence inconsistent data modalities. Then, design Scenarios-Content Introduction Module (SCIM) effectively extract feature synthesized data. Finally, added first stage, features severely pseudo information. Feature Attention (FAM) discriminately select combine network. Extensive experiments on six validate effectiveness method, leading maximum gain 0.439 dB. can even SOTA achieve higher

Language: Английский

Citations

3

KULF-TT53: A Display-Specific Turntable-Based Light Field Dataset for Subjective Quality Assessment DOI Open Access
Kamran Javidi, Maria G. Martini, Peter A. Kara

et al.

Electronics, Journal Year: 2023, Volume and Issue: 12(23), P. 4868 - 4868

Published: Dec. 2, 2023

Light field datasets enable researchers to conduct both objective and subjective quality assessments, which are particularly useful when acquisition equipment or resources not available. Such may vary in terms of capture technology methodology, content, characteristics (e.g., resolution), the availability ratings. When contents a light dataset visualized on display, display system matches received input its output capabilities through various processes, such as interpolation. Therefore, one most straightforward methods create for specific is consider visualization parameters during acquisition. In this paper, we introduce novel display-specific dataset, captured using DSLR camera turntable rig. The visual data seven static scenes were recorded twice by two settings angular resolution. While acquired uniformly within 53-degree angle, viewing cone they for, consists 70 views per while other 140. Capturing was more solution than downsampling, latter approach could either degrade make FOV size inaccurate. paper provides detailed characterization contents, well compressed variations with codecs, together calculated values commonly-used metrics contents. We expect that will be research community working compression, processing, assessment, instance perform assessment tests test new interpolation metrics. future work, also focus provide relevant results. This made free access community.

Language: Английский

Citations

3

Lightweight network with masks for light field image super-resolution based on swin attention DOI
Xingzheng Wang, Shaoyong Wu, Jiahui Li

et al.

Multimedia Tools and Applications, Journal Year: 2024, Volume and Issue: 83(33), P. 79785 - 79804

Published: Feb. 29, 2024

Language: Английский

Citations

0

Depth-Guided Full-Focus Super-Resolution Network for Light Field Images DOI
Deqian Kong, Yan Yuan, Lijuan Su

et al.

Published: Jan. 19, 2024

Light field (LF) imaging system captures the two-dimensional (2D) spatial and 2D angular information of scenes within a single exposure time. Due to this distinctive feature, technique has been rapidly developed over past two decades. However, LF images suffer from low resolution. Currently, numerous deep learning (DL)-based approaches have employed address issue. existing super-resolution (SR) networks ignore defocus blur caused by depth variations, fail yield high-resolution (HR) full-focus directly processing with information. In paper, tackle challenge, we propose new SR method reconstruct HR low-resolution (LR) multi-defocus images. To accomplish task, The degraded dataset is generated utilizing intrinsic as guidance designing spatially-variable (SV) degradation method. designed parts: depth-guided image partitioning process degradation-prior-SR network. Experimental results indicated that our outperforms other both quantitatively qualitatively.

Language: Английский

Citations

0