Single image dehazing motivated by Retinex theory
Jianjun Zhou, Fugen Zhou · 2013
Haze removal from a degraded image which is attenuated by the turbid media in the atmosphere is an important task in image and video processing. From a computer vision perspective, this task is extremely challenging because the haze is dependent on the unknown depth information, and will be under-constrained if only one single haze image is input. If the depth distribution is known, the problem can be well solved. In this paper, we propose a new method to estimate the transmission from which depth distribution can be obtained. It is based on the assumption that the airlight which depends on the distance of objects to the observers, tends to be smooth, and is similar to the Retinex theory. Motivated by this, we use a simple Gaussian filter to convolve the image to obtain the airlight just like we did in Retinex. Experimental results verify the effectiveness of this method.