一种基于反转融合框架的图像曝光校正方法

Image Exposure Correction Method Based on Inversion Fusion Framework

针对单幅图像中存在非正常曝光的问题,在去雾模型能够有效解决图像曝光校正问题的理论指导下,对去雾模型中的透射率进行改进,提出了一种基于反转融合框架的图像曝光校正方法。首先,对过度曝光的局部高亮光源进行雾度建模,采用改进去雾模型完成过度曝光校正任务;针对曝光不足校正问题,通过反转操作得到伪雾图像。然后联合去雾模型及视网膜大脑皮层(Retinex)理论和去雾方法间的对偶性得到曝光不足区域校正的结果图像。最后,借助多尺度图像融合技术生成新的金字塔权重图,利用拉普拉斯金字塔重建图像得到最终的校正结果。将所提方法与四种主流图像校正方法进行对比,结果表明,该方法能够有效解决单幅图像中的非正常曝光问题,并且最大限度减少图像失真、光晕伪影等因素的干扰。

It has been theoretically proven that dehazing models can effectively solve image exposure correction. To solve the abnormal exposure in a single image, we improved the transmittance in the dehazing model and proposed an image exposure correction method using the inversion fusion framework. First, we performed haze modeling for the overexposed high-intensity light source in the local area. Thereafter, we used the improved dehazing model to complete the overexposure correction task. For the underexposure correction problem, we obtained the pseudo-haze image using the inversion operation. The underexposure correction result image was obtained by combining the dehazing model and duality formula between the Retinex theory and dehazing method. Finally, we generated a new pyramid weight map using multiscale image fusion technology, and the final correction result was obtained via Laplacian pyramid reconstruction. Furthermore, we compared the proposed method with four mainstream image correction methods. The experimental results show that the proposed method corrects the abnormal exposure areas of a single image and minimizes the interference image distortion and halo artifacts.