HistoHDR-Net: Histogram Equalization for Single LDR to HDR Image Translation

TL;DR

This work tackles single-image HDR reconstruction by leveraging histogram-equalized LDR inputs. It introduces HistoHDR-Net, a dual-branch ResNet50 architecture that fuses features from the original LDR and histogram-equalized LDR, enhanced by a self-attention module before HDR decoding. A novel five-term loss combines $\mathcal{L}_{L1}$, $\mathcal{L}_{VGG}$, Weber's-law based $\mathcal{L}_{W}$, MS-SSIM-based $\mathcal{L}_{SIM}$, and $\mathcal{L}_{C}$ from $\Delta E^*$ color differences, with tone-mapped supervision via $T(\cdot)$ and $\mu=5000$. The method achieves state-of-the-art performance on SSIM and HDR-VDP-2 in within-dataset and comparable results across datasets, highlighting the effectiveness of histogram-based fusion, self-attention, and perceptual-aware loss for HDR reconstruction.

Abstract

High Dynamic Range (HDR) imaging aims to replicate the high visual quality and clarity of real-world scenes. Due to the high costs associated with HDR imaging, the literature offers various data-driven methods for HDR image reconstruction from Low Dynamic Range (LDR) counterparts. A common limitation of these approaches is missing details in regions of the reconstructed HDR images, which are over- or under-exposed in the input LDR images. To this end, we propose a simple and effective method, HistoHDR-Net, to recover the fine details (e.g., color, contrast, saturation, and brightness) of HDR images via a fusion-based approach utilizing histogram-equalized LDR images along with self-attention guidance. Our experiments demonstrate the efficacy of the proposed approach over the state-of-art methods.

Figures (4)

• Figure 1: Our method HistoHDR-Net (right) can recover the text on the notice board better than the state-of-the-art barua2023arthdr (middle) given an extremely over-exposed LDR image (left) as input.
• Figure 2: Architecture of the proposed HistoHDR-Net method.
• Figure 3: Original and histogram-equalized LDR images. The green and orange histograms are for the original $\text{LDR}_{\text{GT}}$ and histogram-equalized $\text{LDR}_{\text{His}}$ images, respectively.
• Figure 4: HDR images generated by the proposed HistoHDR-Net and state-of-the-art methods.