TY - GEN
T1 - Depth Order Aware Penalization of False EPI Line Response on Light Field View Synthesis
AU - Yim, Jonghoon
AU - Jeon, Byeungwoo
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - To overcome the inherent trade-off between the spatial and angular resolutions of light field (LF) images, much research has been carried out on enhancing the angular resolution of an LF image via a view synthesis. Since the height of an Epipolar Plane Image (EPI) corresponds to its angular resolution, LF view synthesis with a higher angular resolution can be performed by stretching its height. Among the existing LF view synthesis research, there is an approach entailing the line detection and line matching processes in EPI domain. However, detecting lines in an EPI of a low angular resolution can come with fake lines, which results in unpleasant artifacts on the synthesized views. To avoid the problem, we present a method of penalizing the response of fake lines in EPI domain based on the possible occlusion cases. Our test results on synthetic and real-world LF datasets show that our method can respectively achieve PSNR and SSIM around 38.67dB / 0.9507 and 40.89dB / 0.9835 on the synthesized views.
AB - To overcome the inherent trade-off between the spatial and angular resolutions of light field (LF) images, much research has been carried out on enhancing the angular resolution of an LF image via a view synthesis. Since the height of an Epipolar Plane Image (EPI) corresponds to its angular resolution, LF view synthesis with a higher angular resolution can be performed by stretching its height. Among the existing LF view synthesis research, there is an approach entailing the line detection and line matching processes in EPI domain. However, detecting lines in an EPI of a low angular resolution can come with fake lines, which results in unpleasant artifacts on the synthesized views. To avoid the problem, we present a method of penalizing the response of fake lines in EPI domain based on the possible occlusion cases. Our test results on synthetic and real-world LF datasets show that our method can respectively achieve PSNR and SSIM around 38.67dB / 0.9507 and 40.89dB / 0.9835 on the synthesized views.
KW - Depth
KW - Epipolar Plane Image
KW - Light Field
KW - Line Detection
KW - Penalization
KW - View Synthesis
UR - https://www.scopus.com/pages/publications/105016413292
U2 - 10.1109/ITC-CSCC66376.2025.11137668
DO - 10.1109/ITC-CSCC66376.2025.11137668
M3 - Conference contribution
AN - SCOPUS:105016413292
T3 - 2025 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2025
BT - 2025 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2025
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2025 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2025
Y2 - 7 July 2025 through 10 July 2025
ER -