基于局部细节增强的快速多曝光图像融合算法

Fast Multi-exposure Image Fusion Algorithm Based on Local Detail Enhancement

当前基于细节增强机制的多曝光图像融合方法效率较低,针对此问题,提出一种基于局部细节增强的快速多曝光图像融合算法。首先,改进每幅图像待增强区域的划定方法;然后,采用快速局部拉普拉斯滤波(FLLF)提取待增强区域的细节信息,并根据采样插值原理对增强过程进行优化加速,减少计算局部拉普拉斯金字塔系数时的采样次数,避免细节过度增强的同时,进一步提高融合效率。对40组多曝光序列的融合质量和平均运行时间进行测试,并与5种代表性算法进行对比实验。结果表明,算法不仅保持超亮超暗区域的细节信息,而且视觉效果更加自然。算法的两种客观质量指标MEF-SSIM值和Q CB分别为0.942和0.441,在6种算法中分别排名第2和第1;运行时间比基于原始分辨率空间运算的最快方法Li2020减少了46.2%,具有较高的运行效率。

A fast multi-exposure image fusion algorithm based on local detail enhancement is proposed to address the low efficiency of current multi-exposure image fusion methods based on detail enhancement mechanism.Firstly,the method for defining regions of enhancement in each image is improved.Then,fast local Laplacian filtering(FLLF)is used to extract the detail information of the regions to be enhanced,and the enhancement process is optimized and accelerated according to the principle of sampling interpolation,reducing the sampling times of computing local Laplacian pyramid coefficients to avoid excessive enhancement of details and improving the fusion efficiency further.By testing the fusion quality and average running time of 40 sets of multi-exposure sequences and comparing with 5 representative algorithms,the results indicate that the algorithm not only maintains the detailed information of ultra-bright and ultra-dark areas,but also produces a more natural visual effect.The two average objective quality indicators of algorithm,MEF-SSIM values and Q CB,are 0.942 and 0.441 respectively,ranking second of six and first of six.The running time is reduced by 46.2%compared to the fastest method Li2020 based on original resolution space operation,which has higher running efficiency.