Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Electrical and Electronic Engineering | en_US |
| dc.contributor.advisor | Wang, Yi (EEE) | en_US |
| dc.creator | Xie, Yubo | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13917 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | A hybrid deep neural network for raw image super-resolution | en_US |
| dcterms.abstract | In recent years, image super-resolution (SR) techniques based on deep learning have become an important research direction in the fields of computer vision and image processing, and have made significant progress in terms of algorithm performance and computational efficiency. Existing research mainly focuses on image super-resolution reconstruction in RGB colour space by constructing a synthetic dataset containing degradation models (such as downsampling blur, additive noise, etc.), and using deep neural networks (such as convolutional neural networks, etc.) to learn the end-to-end mapping from a low-resolution (LR) image to a high-resolution (HR) image. end-to-end mapping. | en_US |
| dcterms.abstract | However, these methods typically assume that the input is an RGB image processed through the Image Signal Processing (ISP) pipeline, ignoring the characteristics of the sensor’s raw RAW data. Since RAW images retain unprocessed radiometric information with higher bit depth, their super-resolution reconstruction has unique advantages in improving image quality and enhancing the performance of the downstream ISP pipeline, but super-resolution research in the RAW domain is still in the preliminary stage of exploration. In this paper, the key issues and methods of super-resolution for RAW images are investigated. Firstly, a degradation model applicable to the RAW domain is demonstrated; secondly, a synthetic training dataset is constructed based on the model, and a neural network architecture dedicated to RAW data processing is designed, focusing on cross-channel information fusion and noise robustness optimisation; lastly, the superiority of the proposed method in terms of quantitative metrics (e.g., PSNR, SSIM) and visual quality is verified through comparative experiments. The experimental results show that compared with the traditional RGB domain super-resolution method, the direct processing of RAW data can effectively preserve more high-frequency details and provide higher-quality inputs for the subsequent ISP process. This study provides a feasible technical solution for RAW image super-resolution, as well as new ideas for the optimisation of end-to-end computational photography systems. | en_US |
| dcterms.extent | viii, 58 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2025 | en_US |
| dcterms.educationalLevel | M.Sc. | en_US |
| dcterms.educationalLevel | All Master | en_US |
| dcterms.accessRights | restricted access | en_US |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8326.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 4.8 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13917