Fidelity-preserving Learning-Based Image Compression: Loss Function and Subjective Evaluation Methodology
Shima Mohammadi, Yaojun Wu, João Ascenso
TL;DR
The paper tackles fidelity preservation in learning-based image compression by introducing a perceptually optimized loss that combines LPIPS and GAN adversarial loss, implemented in two codecs (LBIC-CO and LBIC-PO). A novel JPEG AIC-2–based triplet subjective methodology evaluates fidelity beyond traditional perceptual appeal, enabling fidelity-focused comparisons. Empirical results show LBIC-PO yields perceptual fidelity gains at low bitrates, with diminishing or negligible benefits at higher bitrates, supporting the viability of fidelity-aware optimization for bandwidth-constrained scenarios. The work also provides a crowdsourced subjective dataset and a framework for integrating perceptual fidelity considerations into JPEG AI verification and encoder-side decision-making.
Abstract
Learning-based image compression methods have emerged as state-of-the-art, showcasing higher performance compared to conventional compression solutions. These data-driven approaches aim to learn the parameters of a neural network model through iterative training on large amounts of data. The optimization process typically involves minimizing the distortion between the decoded and the original ground truth images. This paper focuses on perceptual optimization of learning-based image compression solutions and proposes: i) novel loss function to be used during training and ii) novel subjective test methodology that aims to evaluate the decoded image fidelity. According to experimental results from the subjective test taken with the new methodology, the optimization procedure can enhance image quality for low-rates while offering no advantage for high-rates.