Learned HDR Image Compression for Perceptually Optimal Storage and Display

TL;DR

The paper tackles efficient HDR image compression that preserves perceptual quality for both storage and display. It introduces EPIC-HDR, an end-to-end learned coder that splits HDR into an LDR base stream for display compatibility and an HDR side-information stream for reconstruction, optimized with perceptual rate-distortion using $d_H$ and NLPD losses. The rate function combines $r_H$ and $r_L$ estimated by a hyper-prior+autoregressive entropy model, and the overall loss is $\\ell = r_\mathrm{H} + \lambda_\mathrm{H}d_\mathrm{H} + r_\mathrm{L} + \lambda_\mathrm{L}d_\mathrm{L}$. An automated extension using multi-exposure fusion generates a stack of LDRs conditioned on four luminances and fuses them for improved performance. Experimental results on panoramic HDR data demonstrate superior HDR and LDR RD performance over nine baselines, validating the perceptual optimization approach and the practicality of the two-stream design.

Abstract

High dynamic range (HDR) capture and display have seen significant growth in popularity driven by the advancements in technology and increasing consumer demand for superior image quality. As a result, HDR image compression is crucial to fully realize the benefits of HDR imaging without suffering from large file sizes and inefficient data handling. Conventionally, this is achieved by introducing a residual/gain map as additional metadata to bridge the gap between HDR and low dynamic range (LDR) images, making the former compatible with LDR image codecs but offering suboptimal rate-distortion performance. In this work, we initiate efforts towards end-to-end optimized HDR image compression for perceptually optimal storage and display. Specifically, we learn to compress an HDR image into two bitstreams: one for generating an LDR image to ensure compatibility with legacy LDR displays, and another as side information to aid HDR image reconstruction from the output LDR image. To measure the perceptual quality of output HDR and LDR images, we use two recently proposed image distortion metrics, both validated against human perceptual data of image quality and with reference to the uncompressed HDR image. Through end-to-end optimization for rate-distortion performance, our method dramatically improves HDR and LDR image quality at all bit rates.

Figures (12)

• Figure 1: Comparison between (a) the traditional manually designed and (b) the proposed learned HDR image compression systems.
• Figure 2: System diagram of EPIC-HDR. $\boldsymbol{l}_a(\cdot)$ and $\boldsymbol{h}_a(\cdot)$ are analysis networks, while $\boldsymbol{l}_s(\cdot)$ and $\boldsymbol{h}_s(\cdot)$ are synthesis networks.
• Figure 3: Computational structure of the entropy model. $\boldsymbol{g}_a$ is the hyper-analysis network and $\boldsymbol{g}_s$ is the hyper-synthesis network. The entropy network is used to estimate Gaussian parameters (i.e., the means and variances).
• Figure 4: (a)-(d) Stack of four pseudo-multi-exposure LDR images generated with different maximum luminances $\{10^{4}, 10^{5}, 10^{6}, 10^{7}\}$$\rm cd/m^{2}$, and (e) the corresponding fused LDR image by Li2020fast.
• Figure 5: HDR rate-distortion performance evaluation.