Skip to Main content Skip to Navigation
Theses

A compact video representation format based on spatio-temporal linear embedding and epitome

Martin Alain 1
1 Sirocco - Analysis representation, compression and communication of visual data
IRISA-D5 - SIGNAUX ET IMAGES NUMÉRIQUES, ROBOTIQUE, Inria Rennes – Bretagne Atlantique
Abstract : Efficient video compression is nowadays a critical issue, and is expected to be more and more crucial in the future, with the ever increasing video traffic and the production of new digital video formats with high resolution, wide color gamut, high dynamic range, or high frame rate. The MPEG standard HEVC is currently one of the most efficient video compression scheme, however, addressing the future needs calls for novel and disruptive methods. In fact, the main principles of modern video compression standards rely on concepts designed more than 30 years ago: the reduction of spatial and temporal redundancies, through prediction tools, the use of a transform to further reduce the inner correlations of the signal, followed by quantization to remove non-perceptive information, and entropy coding to remove the remaining statistical redundancies. In this thesis, we explore novel methods which aims at further exploiting the natural redundancies occurring in video signals, notably through the use of multi-patches techniques. First, we introduce LLE-based multi-patches methods in order to improve Inter prediction, which are then combined for both Intra and Inter predictions, and are proven efficient over H.264. We then propose epitome-based de-noising methods to improve the performances of existing codecs in a out-of-the-loop scheme. High quality epitomes are transmitted to the decoder in addition to the coded sequence, and we can then use at the decoder side multi-patches de-noising methods relying on the high quality patches from the epitomes, in order to improve the quality of the decoded sequence. This scheme is shown efficient compared to SHVC. Finally, we proposed another out-of-the-loop scheme relying on a symmetric clustering of the patches performed at both encoder and decoder sides. At the encoder side, linear mappings are learned for each cluster between the coded/decoded patches and the corresponding source patches. The linear mappings are then sent to the decoder and applied to the decoded patches in order to improve the quality of the decoded sequence. The proposed scheme improves the performances of HEVC, and is shown promising for scalable schemes such as SHVC.
Complete list of metadatas

https://hal.inria.fr/tel-01261590
Contributor : Abes Star :  Contact
Submitted on : Monday, March 7, 2016 - 2:52:15 PM
Last modification on : Friday, July 10, 2020 - 4:07:44 PM
Long-term archiving on: : Sunday, November 13, 2016 - 8:35:04 AM

File

ALAIN_Martin.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01261590, version 2

Citation

Martin Alain. A compact video representation format based on spatio-temporal linear embedding and epitome. Signal and Image processing. Université Rennes 1, 2016. English. ⟨NNT : 2016REN1S001⟩. ⟨tel-01261590v2⟩

Share

Metrics

Record views

519

Files downloads

895