Learning a Reinforced Agent for Flexible Exposure Bracketing Selection

TL;DR

This paper introduces EBSNet, a reinforcement learning-based neural network that automatically selects exposure bracketing for high dynamic range imaging without requiring extensive camera information or multiple previews.

Contribution

The authors propose a novel deep neural network, EBSNet, that uses reinforcement learning to select optimal exposure settings from a single preview image, eliminating previous restrictions.

Findings

EBSNet outperforms recent state-of-the-art methods in exposure selection.

Joint training of EBSNet and MEFNet improves multi-exposure fusion quality.

A new benchmark dataset is provided for future research in exposure bracketing.

Abstract

Automatically selecting exposure bracketing (images exposed differently) is important to obtain a high dynamic range image by using multi-exposure fusion. Unlike previous methods that have many restrictions such as requiring camera response function, sensor noise model, and a stream of preview images with different exposures (not accessible in some scenarios e.g. some mobile applications), we propose a novel deep neural network to automatically select exposure bracketing, named EBSNet, which is sufficiently flexible without having the above restrictions. EBSNet is formulated as a reinforced agent that is trained by maximizing rewards provided by a multi-exposure fusion network (MEFNet). By utilizing the illumination and semantic information extracted from just a single auto-exposure preview image, EBSNet can select an optimal exposure bracketing for multi-exposure fusion. EBSNet and MEFNet can be jointly trained to produce favorable results against recent state-of-the-art approaches. To facilitate future research, we provide a new benchmark dataset for multi-exposure selection and fusion.