[Paper Review] A Bio-Inspired Multi-Exposure Fusion Framework for Low-light Image Enhancement

Zhenqiang Ying, Ge Li+1 more|arXiv (Cornell University)|Nov 2, 2017

Image Enhancement Techniques36 references248 citations

TL;DR

The paper presents a bio-inspired multi-exposure fusion framework for low-light enhancement, introducing a dual-exposure fusion algorithm that estimates illumination, synthesizes a well-exposed image via a camera response model, and fuses it with the input image using a learned weight map to reduce contrast and lightness distortion.

ABSTRACT

Low-light images are not conducive to human observation and computer vision algorithms due to their low visibility. Although many image enhancement techniques have been proposed to solve this problem, existing methods inevitably introduce contrast under- and over-enhancement. Inspired by human visual system, we design a multi-exposure fusion framework for low-light image enhancement. Based on the framework, we propose a dual-exposure fusion algorithm to provide an accurate contrast and lightness enhancement. Specifically, we first design the weight matrix for image fusion using illumination estimation techniques. Then we introduce our camera response model to synthesize multi-exposure images. Next, we find the best exposure ratio so that the synthetic image is well-exposed in the regions where the original image is under-exposed. Finally, the enhanced result is obtained by fusing the input image and the synthetic image according to the weight matrix. Experiments show that our method can obtain results with less contrast and lightness distortion compared to that of several state-of-the-art methods.

Motivation & Objective

• Motivate improved low-light image enhancement that preserves natural lightness and contrast.
• Propose a framework inspired by human visual system to generate and fuse multi-exposure images from a single input.
• Develop a dual-exposure fusion algorithm with illumination-based weighting and a camera response model.
• Demonstrate reduced contrast and lightness distortion compared with state-of-the-art methods across multiple datasets.

Proposed method

• Introduce a four-part framework: Multi-Exposure Sampler, Generator, Evaluator, and Combiner.
• Estimate a scene illumination map T by solving a regularized optimization to refine a local-consistent illumination.
• Model a Brightness Transform Function (BTF) g using a two-parameter form g(P,k)=β P^γ and derive a camera response function (CRF) f from a comparametric equation, yielding g(P,k)=e^{b(1-k^{a})} P^{k^{a}} with parameters derived from camera characteristics.
• Compute an optimal exposure ratio k by maximizing the entropy of the brightness channel after synthesis, and fuse the input image with the synthetic well-exposed image using a weight map Ŵ.
• Provide a closed-form illumination map solution via a linear system for efficiency, and validate through extensive experiments on public datasets.

Experimental results

Research questions

• RQ1 Can a bio-inspired multi-exposure fusion framework improve low-light image enhancement by balancing exposure across regions?
• RQ2 How can illumination estimation and a camera response model be combined to synthesize a well-exposed image from a single low-light input?
• RQ3 Does dual-exposure fusion with an illumination-guided weight map reduce contrast and lightness distortion compared to existing methods?

Key findings

• The proposed Ours method yields lower lightness distortion (LOE) across multiple datasets compared to MSRCR, Dong, NPE, LIME, MF, and SRIE.
• In terms of Visual Information Fidelity (VIF), the proposed method achieves higher scores on several datasets, indicating better preservation of visual information.
• The method demonstrates favorable contrast preservation with lower distortion as measured by DRIM across datasets.
• The approach balances enhancement and naturalness, reducing halo artifacts and noise compared with several baselines.
• The authors report competitive time costs and provide open-source code for reproducibility.