基于偏振测量的雾天图像场景分割

现有场景分割方法主要依赖于图像亮度、颜色和纹理等特征,然而在雾天图像中提取这些特征将变得困难且不稳定.基于此本文提出了适用于雾天图像场景分割的特征矢量,以及相应的特征提取算法.特征矢量由目标偏振度、深度和颜色三部分组成.特征提取算法分别为:用去相关的方法从图像偏振度分离出大气偏振度和目标偏振度;根据雾天退化模型和雾天图像偏振表示形式推导出场景深度信息;利用两幅偏振图像求出非偏振彩色图像,从而得到场景的颜色信息.将这些特征构成的特征矢量用于基于图的分割算法中,并从两个方面比较了仅使用颜色特征和使用本文特征矢量的分割结果.最后得出结论:对于雾天图像而言,这些特征比通常的颜色特征更加有效和鲁棒.

Scene Segmentation of Hazy Image Using Polarization Measurements

The general segmentation algorithms are mainly dependent on visual features extracted from images, such as intensity, color and texture features. However, the feature extraction from hazy image becomes difficult and unstable. In this paper, a novel feature vector suitable to hazy image segmentation is introduced, and the corresponding feature extraction algorithm is proposed. The feature vector is composed of three components, which are degree of polarization of objects, depth map and color information, respectively. The degree of objects is separated with the degree of airlight using an decorrelation-based method. The depth map is inferred by analyzing the polarization images based on the hazy degradation model. And the color information is obtained from non-polarization image estimated by polarization images. Finally, these features are formed into a vector and fed to a well-known graph-based segmentation algorithm. After comparing the proposed results with those using only color cue on two aspects, it can be concluded that the proposed feature vector is more effective and robust than the usual features for hazy scene segmentation.