The Library
Graph based transforms for block-based predictive transform coding
Tools
Roy, Debaleena (2023) Graph based transforms for block-based predictive transform coding. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Roy_2023.pdf - Submitted Version - Requires a PDF viewer. Download (19Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3985230
Abstract
Orthogonal transforms are the key aspects of the encoding and decoding process in many state-of-the-art compression systems. The transforms in blockbased predictive transform coding (PTC) is essential for improving coding performance, as it allows decorrelating the signal in the form of transform coefficients. Recently, the Graph-Based Transform (GBT), has been shown to attain promising results for data decorrelation and energy compaction especially for block-based PTC. However, in order to reconstruct a frame for GBT using block-based PTC, extra-information is needed to be signalled into the bitstream, which may lead to an increased overhead. Additionally, the same graph should be available at the reconstruction stage to compute the inverse GBT of each block.
In this thesis, we propose a set of a novel class of GBTs to enhance the performance of transform. These GBTs adopt several methods to address the issue of the availability of the same graph at the decoder while reconstructing video frames. Our methods to predict the graph can be categorized in two types: non-learning-based approaches and deep learning (DL) based prediction. For the first type our method uses reference samples and template-based strategies for reconstructing the same graph. For our next strategies we learn the graphs so that the information needed to compute the inverse transform is common knowledge between the compression and reconstruction processes. Finally, we train our model online to avoid the amount, quality, and relevance of the training data.
Our evaluation is based on all the possible classes of HEVC videos, consist of class A to F/Screen content based on their varied resolution and characteristics. Our experimental results show that the proposed transforms outperforms the other non-trainable transforms, such as DCT and DCT/DST, which are commonly employed in current video codecs in terms of compression and reconstruction quality.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software | ||||
Library of Congress Subject Headings (LCSH): | Image compression -- Standards, Video compression, Neural networks (Computer science), Diagnostic imaging -- Data processing, Image processing | ||||
Official Date: | January 2023 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Department of Computer Science | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Sanchez, Victor ; Guha, Tanaya | ||||
Format of File: | |||||
Extent: | xv, 126 pages : illustrations | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |