A haze density aware adaptive perceptual single image haze removal algorithm

Haze or fog jeopardizes both environment and image quality, which degrades the quality of subsequent computer vision algorithms. Recently haze removal method in image processing makes significant progress. The existing methods usually require complicated manual parameters setting according to the variance of input. Among them, dehazing method based on dark channel prior is considered to be the most efficient one. However, the problems brought by dark channel prior method including low luminance, sky region distortion and low saturation are inevitable. The proposed dehaze method in this paper can adaptively adjust parameters settings by introducing haze density detection. Besides, the proposed method improves the original dim recovered image by adaptively adjust exposure and color saturation in YCbCr color space. Furthermore, fast guided filter is employed to refine the transmission map. The experimental results show that the proposed method performs better both objectively and subjectively.