Single Image Dehazing with White Balance Correction and Image Decomposition

Single image dehazing has been a challenging problem due to its ill-posed nature. While most of the existing single image based dehazing algorithms address this issue by introducing certain assumptions and priors into the haze imaging model, the imaging process of imaging devices has been seldom taken into account, such as white balance and metering. In general, consumer photos are taken with AWB (Auto White Balance). Hence, color temperature in a foggy scenario may not be correctly detected, which results in color distortion; and the whole scene looks brighter, which leads to under-exposure during the imaging process. In this paper, we propose to handle these two issues by applying white balance correction and decomposing an image into two component images, reflex lightness image and ambience illumination image. We devise an improved dark channel prior based algorithm to dehaze the reflex lightness image and the exposure adjustment is estimated from the ambience illumination image. Finally, a high quality haze- free image is produced by refining the brightness of the preliminarily dehazed image with the estimated exposure adjustment. Experimental results with a benchmark dataset demonstrate that our approach outperforms the state-of-the-art, in terms of contrast and color fidelity.