TY - GEN
T1 - 3D Super-resolution Optical Imaging Using Deep Image Prior
AU - Wang, Ruoxin
AU - Cheung, Chi Fai
N1 - Funding Information:
The work described in this paper was mainly supported by a grant from the Research Grants Council of the Government of the Hong Kong Special Administrative Region, China (Project No. 15202717). The authors would also like to express their sincere thanks to the Research Committee of The Hong Kong Polytechnic University for their financial support of the project through a PhD studentship (project account code: RK36).
Publisher Copyright:
© 2021 IEEE.
PY - 2021/8/27
Y1 - 2021/8/27
N2 - Deep learning based super-resolution methods have received much attention, especially unsupervised super-resolution due to the difficulty of collecting images pairs (low-resolution and high-resolution images from the same scenario) in many fields, such as optics. Optical imaging is typical technique in advance optical measurement equipment and optical super-resolution imaging has received much attention. In this paper, a novel model, deep image prior with design surface model (DIP-DSM), based on deep image prior to improve the resolution of optical imaging is presented. It makes use of single image instead of using random input in which the design surface model is regarded as prior information. To validate the model, a series of experiments are conducted, and the results show the superiority of the proposed model as compared with deep image prior. Furthermore, the performance of different neural networks are explored and it is find that the U-Net achieve best reconstruction quality and reach to PSNR, 32.937.
AB - Deep learning based super-resolution methods have received much attention, especially unsupervised super-resolution due to the difficulty of collecting images pairs (low-resolution and high-resolution images from the same scenario) in many fields, such as optics. Optical imaging is typical technique in advance optical measurement equipment and optical super-resolution imaging has received much attention. In this paper, a novel model, deep image prior with design surface model (DIP-DSM), based on deep image prior to improve the resolution of optical imaging is presented. It makes use of single image instead of using random input in which the design surface model is regarded as prior information. To validate the model, a series of experiments are conducted, and the results show the superiority of the proposed model as compared with deep image prior. Furthermore, the performance of different neural networks are explored and it is find that the U-Net achieve best reconstruction quality and reach to PSNR, 32.937.
KW - Deep image prior
KW - deep learning
KW - measurement
KW - optical imaging
KW - precision metrology
KW - precision surface measurement
KW - Super-resolutuon
UR - http://www.scopus.com/inward/record.url?scp=85115445116&partnerID=8YFLogxK
U2 - 10.1109/ICOIM52180.2021.9524418
DO - 10.1109/ICOIM52180.2021.9524418
M3 - Conference article published in proceeding or book
AN - SCOPUS:85115445116
T3 - 2021 International Conference of Optical Imaging and Measurement, ICOIM 2021
SP - 5
EP - 8
BT - 2021 International Conference of Optical Imaging and Measurement, ICOIM 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2021 International Conference of Optical Imaging and Measurement, ICOIM 2021
Y2 - 27 August 2021 through 29 August 2021
ER -