Haze (or fog, mist, and other atmospheric phenomena) is a main degradation of outdoor images, weakening both colors and contrasts. We propose a simple but effective "dark channel prior" to remove haze from a single input image. The dark channel prior is a kind of statistics of outdoor haze-free images. It is based on a key observation - most local patches in outdoor haze-free images contain some pixels whose intensity is very low in at least one color channel. Using this prior with the haze imaging model, we can directly estimate the thickness of the haze and recover a high quality haze-free image. Results on a variety of hazy images demonstrate the power of the proposed prior. Moreover, a high quality depth map can also be obtained as a by-product of haze removal.

our result

our recovered depth

our result

our recovered depth

Related publications:

• Single Image Haze Removal using Dark Channel Prior, by Kaiming He, Jian Sun, and Xiaoou Tang, in CVPR 2009 (Oral, Best Paper Award).

• Guided Image Filtering, by Kaiming He, Jian Sun, and Xiaoou Tang, in ECCV 2010 (Oral).

• Single Image Haze Removal using Dark Channel Prior, by Kaiming He, Jian Sun, and Xiaoou Tang, in TPAMI 2011 (Spotlight Paper).

• Guided Image Filtering, by Kaiming He, Jian Sun, and Xiaoou Tang, in TPAMI 2013.

Other materials:

• CVPR 2009 presentation slides

• ECCV 2010 presentation slides

• Video results

Result pages:

• More results

• comparisons

Performance    Rather than solve the "soft matting" as in our original CVPR 2009 paper, we have proposed the "guided filter" in our ECCV 2010 paper to efficiently refine the transmission map. The entire dehazing procedure in our latest unoptimized CPU implementation takes about 0.2s per Mp without any downsampling. A faster coarse-to-fine solution can dehaze 32Mp in 0.5s with almost no visual degradation. The performance is reported with an i7 quad-core 3.0Hz CPU, 8G RAM, and 64-bit Win 7.