Modulo video recovery with deep learning
Modulo images are a particular class of images captured by modulo cameras that enable the recovery of theoretically infinite dynamic range images. The basic principle of this image is to perform a modulo operation on values that exceed the maximum dynamic range of the image. Thus, within the overexp...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/173523 |
| _version_ | 1811689153867284480 |
|---|---|
| author | Li,Zike |
| author2 | Tay Wee Peng |
| author_facet | Tay Wee Peng Li,Zike |
| author_sort | Li,Zike |
| collection | NTU |
| description | Modulo images are a particular class of images captured by modulo cameras that enable the recovery of theoretically infinite dynamic range images. The basic principle of this image is to perform a modulo operation on values that exceed the maximum dynamic range of the image. Thus, within the overexposure region of a conventional image, the modulo image can still retain some of the information of the image. Traditional modulo image restoration methods based on Markov random fields have problems such as poor restoration results and high operating costs. The method based on deep learning neural networks can achieve better recovery results. Still, the technique has some redundancy in the processing of modulo video, making it challenging to recover many modulo video frames efficiently. In this dissertation, we combine deep learning and optical flow methods to propose an architecture capable of reconstructing HDR videos from modulo videos. It has been proved experimentally that the architecture has a robust unwrapping effect and running speed. |
| first_indexed | 2024-10-01T05:43:35Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/173523 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T05:43:35Z |
| publishDate | 2024 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1735232024-02-16T15:42:51Z Modulo video recovery with deep learning Li,Zike Tay Wee Peng School of Electrical and Electronic Engineering wptay@ntu.edu.sg Computer and Information Science Modulo images are a particular class of images captured by modulo cameras that enable the recovery of theoretically infinite dynamic range images. The basic principle of this image is to perform a modulo operation on values that exceed the maximum dynamic range of the image. Thus, within the overexposure region of a conventional image, the modulo image can still retain some of the information of the image. Traditional modulo image restoration methods based on Markov random fields have problems such as poor restoration results and high operating costs. The method based on deep learning neural networks can achieve better recovery results. Still, the technique has some redundancy in the processing of modulo video, making it challenging to recover many modulo video frames efficiently. In this dissertation, we combine deep learning and optical flow methods to propose an architecture capable of reconstructing HDR videos from modulo videos. It has been proved experimentally that the architecture has a robust unwrapping effect and running speed. Master's degree 2024-02-14T03:02:51Z 2024-02-14T03:02:51Z 2024 Thesis-Master by Coursework Li, Z. (2024). Modulo video recovery with deep learning. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/173523 https://hdl.handle.net/10356/173523 en application/pdf Nanyang Technological University |
| spellingShingle | Computer and Information Science Li,Zike Modulo video recovery with deep learning |
| title | Modulo video recovery with deep learning |
| title_full | Modulo video recovery with deep learning |
| title_fullStr | Modulo video recovery with deep learning |
| title_full_unstemmed | Modulo video recovery with deep learning |
| title_short | Modulo video recovery with deep learning |
| title_sort | modulo video recovery with deep learning |
| topic | Computer and Information Science |
| url | https://hdl.handle.net/10356/173523 |
| work_keys_str_mv | AT lizike modulovideorecoverywithdeeplearning |