Fast image dehazing using joint Local Linear sure-based filter and image fusion

The method of using dark channel prior to removing haze from a single input image is simple and effective. However, the method of soft matting algorithm with high computational complexity and time consumption can´t meet the demand of real-time performance. In this paper, we present a novel fast image dehazing method for single image based on joint LLSURE (Local Linear Stein´s Unbiased Risk Estimate) filter and image fusion. At first, we roughly estimate the transmission map through a dark channel prior, then the raw estimation is refined by a fast and exact linear-time joint LLSURE filter that preserves edges, and then we get the original dehazed image by inverting the atmospheric scattering model. Next, we propose a pseudo-dehazed image basing on the atmospheric scattering model, then it is fused with the original dehazed image to solve the over-dehazing problem. Experiments show that our method is fast with O(N) time which restores the contrast and color of the scene effectively, significantly improves visibility of the image. Compared with the exiting state-of-the-art dehazing algorithms, our method could get much faster speed and better resolution as well as color reproduction.