# Research on optimized multi-exposure image fusion method for improving information entropy in high-brightness region: Based on developing grayscale feature weight matrix

## TL;DR

This paper introduces a new method to enhance image quality in high-brightness areas by fusing multi-exposure images, improving information entropy and mutual information.

## Contribution

A novel multi-exposure image fusion method is proposed, specifically optimized for high-brightness regions using a grayscale feature weight matrix.

## Key findings

- The proposed method increases information entropy in high-brightness regions by 86.49% compared to classical methods.
- Mutual information is improved by 13.88% in six tested scenarios.
- The method uses a grayscale weight matrix and Laplacian pyramid for effective multi-scale fusion.

## Abstract

This study serves as a preliminary work for image measurement, aiming to support image-based analysis or measurement tasks of high-brightness light environments. Overexposure can lead to significant loss of information in high-brightness areas of images. To address this issue, this study focuses on the core task of enhancing image information entropy (EN) and proposes a novel multi-exposure image fusion (MEF) method tailored to the characteristics of high-brightness regions. First, low-, medium-, and high-exposure urban outdoor artificial light at night (ALAN) images were simultaneously captured. Based on the brightness characteristics of illuminated regions, a grayscale value weight matrix oriented towards increasing pixel value gradient information was developed. With this as the primary factor and saturation and contrast as supplementary references, an optimized MEF weighting strategy was proposed. Finally, multi-scale fusion was achieved through the Laplacian pyramid. The experimental results in six different scenario show that compared with the classical MEF method, this method significantly improves the EN of the fused ALAN region by 86.49% and increases the mutual information (MI) by 13.88%. It provides an important preprocessing solution for image analysis and measurement tasks.