A hybrid features learning model for single image haze prediction

In scene dehazing problem, single image haze prediction is one of the most challenging issues. In this paper, we propose a hybrid features learning model (HFLM) for haze prediction. HFLM takes a hazy image as the input, and outputs its medium transmission map that is subsequently used to recover a haze-free image via atmospheric scattering model. There are two main stages in HFLM. The first stage is used to extract haze-related features from haze images. The second stage aims to establish the mapping relationship between features and medium transmission. In the experimental part, we explore the hyper-parameters in support vector and verify the significance of the features selection. Further, we compare our method with other dehazing methods and make a qualitative comparison on synthetic images. Demonstrate our method has more superior performance than the state-of-the-art dehazing methods.