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 | Yang, Yifei | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13892 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Spatial-adaptive feature modulation and Laplacian pyramid hybrid network for bitstream-corrupted image restoration | en_US |
| dcterms.abstract | Despite the emergence of various image restoration methods, there are few techniques available for recovering JPEG images corrupted by bitstream errors. Recently, Transformers have gained significant attention across all fields of artificial intelligence. Their self-attention mechanism is a key factor in their success, but it comes with high computational costs. Therefore, models need to be lightweight by integrating Convolutional Neural Networks (CNNs). | en_US |
| dcterms.abstract | This paper proposes a simple yet effective deep network to address the super-resolution problem of pre-repairing bitstream-corrupted images. To simulate real-world scenarios where images encounter errors, we automatically inject errors to generate corrupted images, mimicking bitstream damage. Once decoded, these damaged images often exhibit color distortion and block displacement. To tackle these issues, we propose a network to repair corrupted images. Although many image super-resolution solutions have been proposed, they typically require high power consumption and memory usage. Our approach optimizes these drawbacks while ensuring restoration quality. Initially, the error-injected images are decoded and input into a pre-repair module. The processed images are then fed into the network for super-resolution, ultimately resulting in higher-resolution images. Specifically, we developed a Spatial Adaptive Feature Modulation (SAFM) mechanism, which is combined with a Laplacian pyramid to form a hybrid network. We merge the input network images and thumbnails into six-channel data, process and convolve them, and then integrate them into the Laplacian pyramid and thumbnail fusion, finally obtaining a high-definition image. Experimental results demonstrate that this network effectively enhances both the resolution and efficiency of the model. | en_US |
| dcterms.extent | iii, 38 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2024 | en_US |
| dcterms.educationalLevel | M.Sc. | en_US |
| dcterms.accessRights | restricted access | en_US |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8299.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 1.2 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/200/13892