Deep Screen Content Image Compression With Palette Context and Hash Attention
IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 16820-16831
DOI: 10.1109/tcsvt.2026.3714976 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Feifeng Wang und 6 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Shanghai Excellent Academic Leaders Program, Shanghai Leading Talent Program of Eastern Talent Plan.
Abstract (Originalsprache)
Most deep learning (DL)-based image codecs, which are primarily designed for natural images, typically focus on reducing local spatial redundancy among adjacent pixels. However, they exhibit two critical limitations when applied to screen content images (SCIs) compression, namely insufficient exploitation of color redundancy inherent in SCIs and ineffective capture of long-range repetitive patterns. To address these drawbacks, we propose a novel DL-based SCI compression framework, termed DSCIC++, which integrates two core modules: a palette-based context embedder (PCE) and a hash-based feature aggregation (HFA) module. Specifically, instead of directly compressing the raw pixel matrix, the PCE first converts SCIs into a compact palette-based representation and then generates informative palette contexts to facilitate effective compression and high-fidelity reconstruction, while the HFA captures globally repetitive patterns in SCIs through efficient hash attention, enabling the network to explicitly model long-range dependencies of repetitive patterns, thereby further improving compression performance. Extensive experimental results on multiple representative SCI datasets demonstrate that DSCIC++ yields a 19.14% BD-rate reduction compared with the state-of-the-art traditional codec VVC-SCC and also outperforms other competing DL-based image compression methods.
Förderung
- National Natural Science Foundation of China
- Shanghai Excellent Academic Leaders Program
- Shanghai Leading Talent Program of Eastern Talent Plan
Zitieren
APA
Wang, F., Tian, Z., Lei, J., Wang, S., Zhu, Y., Teng, Q., & Shen, L. (2026). Deep Screen Content Image Compression With Palette Context and Hash Attention. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 16820-16831. https://doi.org/10.1109/tcsvt.2026.3714976
Vancouver
Wang F, Tian Z, Lei J, Wang S, Zhu Y, Teng Q, et al. Deep Screen Content Image Compression With Palette Context and Hash Attention. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):16820-16831. doi:10.1109/tcsvt.2026.3714976
BibTeX
@article{Wang2026,
author = {Wang, Feifeng and Tian, Zhaoyi and Lei, Jun and Wang, Shiwei and Zhu, Yao and Teng, Qi and Shen, Liquan},
title = {Deep Screen Content Image Compression With Palette Context and Hash Attention},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
year = {2026},
volume = {36},
number = {11},
pages = {16820--16831},
doi = {10.1109/tcsvt.2026.3714976}
}
EUVIMED