摘要
现有的高动态范围成像(HDRI)方法往往需要采集多幅图像,且在图像重建中易出现伪影,降低了图像质量。为此,提出了一种基于伪曝光融合的HDRI方法。该方法可直接从单幅图像中通过人工重映射生成一组多曝光序列;然后根据对比度、曝光度质量评价求取多曝光序列权重图;接着对权重图进行高斯金字塔分解,对多曝光序列进行拉普拉斯金字塔分解;最后,基于多尺度融合生成高质量的HDR图像。实验结果表明:本方法重建的图像在结构相似度(MEF-SSIM)上的平均评分为0.966,相比较IEF和FSPD算法而言,其MEF-SSIM评分分别提高了17.5%和10.0%。
Existing high dynamic range imaging(HDRI) algorithms cause artifacts in the reconstruction process, which deteriorates image quality. To overcome this problem, a pseudo exposure fusion based HDRI method is proposed. The method can directly generate a group of sequence from a single inputting image by artificial remapping. Then, the weighting map of the sequence is obtained via the contrast and exposedness. The weighting map is decomposed by Gaussian pyramid, and the multi-exposure sequence is decomposed by Laplace pyramid. Finally, a high-quality HDR image is generated based on multiscale exposure fusion. Experimental results show that, the average score of the proposed method on MEF-SSIM is 0.966. Compared with IEF and FSPD algorithms, MEF-SSIM scores are improved by 17.5% and 10.0%, respectively.
作者
吴玲风
胡骏保
李娜
WU Lingfeng;HU Junbao;LI Na(Guangdong University of Science and Technology,Dongguan Guangdong 523083,China;Shenzhen University,Shenzhen Guangdong 518060,China;Xi'an University of Posts and Telecommunications,Xi'an 710121,China)
出处
《激光杂志》
CAS
北大核心
2022年第10期72-77,共6页
Laser Journal
基金
国家自然科学基金项目(No.41874173)
广东科技学院青年项目(No.GKY-2021KYQNK-1)。
