icon

Frequency-Aware Perceptual Optimization for Low-Complexity Implicit Image Compression

Haotian Wu1,2, Gen Li2, Di You2, Pier Luigi Dragotti2, Deniz Gündüz2

1. Zhejiang University 2. Imperial College London

International Conference on Machine Learning (ICML) 2026

Abstract

We propose a frequency-aware perceptual optimization framework for low-complexity image compression, realized as a Realism-enhanced Region-based Implicit Codec (Re2IC). Re2IC models visual perception via saliency-guided region partitioning and local–global perceptual modulation. To enhance realism under complexity constraints, we introduce wavelet–Wasserstein distortion (WA-WD), a frequency-decomposed perceptual distortion that balances fidelity and realism through subband-aware modeling and provides a more reliable approximation than standard Wasserstein distortion. Together, these designs enable fine-grained spatial–spectral optimization, allowing Re2IC to achieve superior rate–perception trade-offs, outperforming generative codecs such as HiFiC while using less than 1% of their decoding cost. Extensive experiments demonstrate state-of-the-art perceptual performance among overfitted codecs. Beyond compression, WA-WD serves as a standalone, tunable perceptual metric with strong alignment to human preference (Pearson 94.6%, Spearman 92.3%) and competitive performance across multiple IQA benchmarks.

Full comparison figure
HiFiC center image
ReReIC center image
ReReIC
HiFiC

Fig. 1: Interactive comparison between ReReIC and HiFiC.

WA-WD 0.129bpp
C3-WDs 0.130bpp

Fig. 2: Visual perceptual comparison between WA-WD and C3-WDs at similar bitrates.

Illustration of WA-WD

Description of the image

Fig. 3: Paired images are processed by a VGG backbone and DWT, decomposing features into orthogonal subbands for multi-scale WA-WD estimation.

Human rating results and decoder complexity on Kodak

Description of the image

Fig. 4: Left: Evaluation of different methods vs. bit rate. Right: Decoding complexity at the middle bit-rate regime.

Numerical results of ReReIC.

Description of the image

Fig. 5: Rate-distortion and -perception curves on Kodak

Description of the image

Fig. 6: Rate-distortion and -perception curves on CLIC2020

Description of the image

Fig. 7: Evaluation of WA-WD: (a) performance across IQA datasets; (b) human-rating prediction.

BibTeX


  @article{ReReIC,
    title={Frequency-Aware Perceptual Optimization for Low-Complexity Implicit Image Compression},
    author={Haotian Wu, Gen Li, Di You, Pier Luigi Dragotti and Deniz Gündüz},
    journal={Proceedings of the 43rd International Conference on Machine Learning, Seoul, South Korea},
    year={2026}
  }