Multi-Model Local Illumination Compensation for Enhanced Compression Model
IEEE Transactions on Circuits and Systems for Video Technology · 2026-09 · 36 (9) · 12796-12813
DOI: 10.1109/tcsvt.2026.3696032 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-09. Verfasst von Jialin Li und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 51 Literaturangaben. Gefördert von: National Key Research and Development Plan, Central Processing Unit (CPU) Cluster Built by Multimedia Computing and Communication (MCC) Laboratory of Information Science and Technology Institution, Unive….
Abstract (Originalsprache)
Inter prediction is a key technique for reducing temporal redundancy in video coding. However, irregular inter-frame illumination changes are hard to predict, yielding poorly predicted residuals that are difficult to compress. To address this, local illumination compensation (LIC) has been introduced into the next-generation exploratory coding model, which uses a linear model (LM) to approximate and compensate for inter-frame illumination variations, thereby reducing the prediction residual. Due to the simple design of LM and the least squares regression (LSR) solution in ECM, LIC suffers notable limitations, including the limited fitting capacity of the LM and the outlier sensitivity of LSR, which makes the compensation less robust to outliers. To address these limitations, we propose a multi-model local illumination compensation (MM-LIC) framework, which introduces the multi-model paradigm to optimize the model type (MMT) and model solution (MMS) jointly. In MMT, the candidate model types are extended beyond LM by incorporating a non-linear model (NLM) and a gradient linear model (GLM) to better handle diverse illumination variations. In MMS, ridge regression (RR) is further introduced to improve model solution accuracy and stability, especially when handling outliers. Meanwhile, in order to preserve decision accuracy without extra bit overhead, we design separable implicit decision mechanisms for the whole compensation process. The proposed method is implemented in ECM-8.0. Experimental results show that the proposed MM-LIC can achieve, on average, 0.45% and 0.23%, and achieve up to 0.68% and 0.40% BD-rate reduction under LDB and RA configurations, respectively, with only slight increases in time complexity. Furthermore, the proposed scheme also demonstrates its potential in the latest ECM version.
Förderung
- National Key Research and Development Plan
- Central Processing Unit (CPU) Cluster Built by Multimedia Computing and Communication (MCC) Laboratory of Information Science and Technology Institution, Unive…
Zitieren
APA
Li, J., Li, Z., Li, Y., Liu, D., & Li, L. (2026). Multi-Model Local Illumination Compensation for Enhanced Compression Model. IEEE Transactions on Circuits and Systems for Video Technology, 36(9), 12796-12813. https://doi.org/10.1109/tcsvt.2026.3696032
Vancouver
Li J, Li Z, Li Y, Liu D, Li L. Multi-Model Local Illumination Compensation for Enhanced Compression Model. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(9):12796-12813. doi:10.1109/tcsvt.2026.3696032
BibTeX
@article{Li2026,
author = {Li, Jialin and Li, Zhuoyuan and Li, Yao and Liu, Dong and Li, Li},
title = {Multi-Model Local Illumination Compensation for Enhanced Compression Model},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
year = {2026},
volume = {36},
number = {9},
pages = {12796--12813},
doi = {10.1109/tcsvt.2026.3696032}
}
EUVIMED