Evaluation of histogram image defogging methods based on histogram equalization, dark channel prior, and convolutional neural network

Abstract

Air pollution condition is getting worse with the advancement of society development, environmental pollution has gradually intensified, and smog frequently occurs in more and more cities. On foggy day, the saturation and contrast of an image could be low, and colors tend to drift and distortion. As a result, seeking a simple and effective image de-fogging technique is important for the subsequent research. In this study, three existing classical de-fogging algorithms are reproduced: histogram equalization, dark channel prior method, and convolutional neural network. The three de-fogging algorithms were compared respectively under the conditions of thin fog, thick fog, high brightness, and low brightness, so as to analyze their advantages and disadvantages. It is concluded that there is no obvious difference among the three algorithms in the de-fogging effect under the conditions of thick fog and high brightness, but relatively speaking, the de-fogging image generated by the dark channel prior is more real. When the fog is thin, the dark channel prior and convolutional neural network work better. Under the condition of low brightness, the histogram equalization has a better de-fogging effect.