为了获得与用户主观感知相一致的颜色校正算法和对校正结果进行客观评估,本文首先创建了一个针对颜色校正的数据集ICCD(Image Color Correction Database).ICCD数据集中的颜色差异涵盖了多种类型和粒度,其中颜色差异类型包括亮度、色相、饱和度、曝光度、对比度以及RGB中的R和G通道,每类颜色差异包括3个修改粒度.本文挑选了6种具有代表性的颜色校正算法对目标图像进行校正,并通过用户调查获得校正结果图像的主观平均得分值.基于ICCD数据集,本文对6种颜色校正算法的性能进行评估,得出在大多数颜色差异和粒度上,Pitie提出的迭代颜色分布转换算法的校正性能最好,同时具有较好的稳定性.最后,本文对14种图像质量评估方法进行评估,挑选出与已有的评估方法相比与主观感知一致性更好的评估方法.%To achieve objective image correction quality assessment results that are consistent with subjective perception,we create an Image Color Correction Database (ICCD).ICCD contains a variety of types and scales of color difference.The types of color difference include the differences in brightness,hue,saturation,exposure,contrast and R and G channels.Each type has three different scales.We select six state-of-the-art color correction algorithms to perform color correction for each target image.Then we design and conduct user study to get users' Mean Opinion Score (MOS).Based on ICCD,we evaluate the performance of six color correction algorithms.For most of the types and scales of color difference,Pitie's iterative color distribution transfer algorithm performs best.We also evaluate the 14 objective image quality assessment metrics and pick out three better assessment metrics that achieve better consistency with MOS than the existing methods.
展开▼
