LAUSR

In the task of haze scene restoration, dark channel prior is the most convincing theory. Its biggest contribution is to introduce the concept of dark channel map, which is convenient for acquiring initial transmission. Minimum filter is the basic operation for dark channel map. Nevertheless, the disadvantage of minimum filter is to cause edge information jump, which further brings Halo effect. Besides, dark channel prior is invalid for highlight areas such as white objects, and it also causes distortion. In this letter, we design a delicate function to replace minimum filter operation, and its purpose is to fundamentally avoid the errors of edge assignment. The function includes two pieces: power-law transform compresses low-luminance regions, and linear attenuation retains original characteristics of bright parts in the image. Additionally, we use logarithmic compression to simulate dark channel map of haze-free image and receive a polishing transmission. Compared with other state-of-the-art algorithms on both synthetic datasets and real-world images, our algorithm shows favorable performance for haze removal.