Faithful and Realistic Image Compression at Extreme-Low Bitrate via Pretrained Discrete Priors
IEEE Signal Processing Letters · 2026 · 33 · 1901-1905
DOI: 10.1109/lsp.2026.3689377 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Min Woo Kim und 1 weiteren Autor:innen. Gefördert von: BK21 FOUR Program of the Education and Research Program, Seoul National University, National Research Foundation.
Abstract (Originalsprache)
Image compression is fundamentally governed by the distortion-perception trade-off, where achieving high perceptual realism often necessitates a sacrifice in pixel-wise fidelity. This challenge is significantly magnified at extreme low bitrates due to severe information constraints, creating a critical bottleneck for data representation. To address this representation bottleneck, we propose a framework that leverages pretrained discrete priors derived from large-scale vision encoders. Rather than training from scratch, we employ clustering-based initialization to map high-capacity discrete priors into a bitrate-constrained space. This ensures that reconstructed images achieve enhanced content fidelity while preserving superior perceptual realism. Furthermore, to overcome the practical bitrate overhead where arithmetic coding occasionally exceeds the theoretical uniform bound in discrete latent spaces, we introduce a dynamic entropy coding scheme. By deterministically switching between arithmetic and uniform coding on a per-image basis, our method ensures near-optimal compression efficiency. Extensive experiments confirm that our method improves content preservation while maintaining high-fidelity perceptual quality.
Förderung
- BK21 FOUR Program of the Education and Research Program
- Seoul National University
- National Research Foundation
- Korean Government
Zitieren
APA
Kim, M. W., & Cho, N. I. (2026). Faithful and Realistic Image Compression at Extreme-Low Bitrate via Pretrained Discrete Priors. IEEE Signal Processing Letters, 33, 1901-1905. https://doi.org/10.1109/lsp.2026.3689377
Vancouver
Kim MW, Cho NI. Faithful and Realistic Image Compression at Extreme-Low Bitrate via Pretrained Discrete Priors. IEEE Signal Processing Letters. 2026;33:1901-1905. doi:10.1109/lsp.2026.3689377
BibTeX
@article{Kim2026,
author = {Kim, Min Woo and Cho, Nam Ik},
title = {Faithful and Realistic Image Compression at Extreme-Low Bitrate via Pretrained Discrete Priors},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {1901--1905},
doi = {10.1109/lsp.2026.3689377}
}
EUVIMED