Traffic image dehazing based on wavelength related physical imaging model

Abstract

Fog concentration is of crucial importance for traffic image dehazing. However, existing methods neglect it, resulting in inferior dehazing results. To tackle this problem, we propose a dehazing method based on the wavelength related physical imaging model. First, we present a wavelength related physical imaging model for traffic imaging. Then, according to the prior that colors of objects in an image are decided based the reflection of different wavelengths, we design a transmission estimation strategy based on the maximal fuzzy correlation and graph cut result. To be specific, the segmentation can be obtained by the maximal fuzzy correlation and the graph cut algorithm, and then it can be used as the guided image in the guided filter for getting a transmission map with continuous scene information. Such a segmentation design can take advantage of both threshold-based segmentation methods, e.g., maximal fuzzy correlation, and spatial-based segmentation methods, e.g., graph cut. Furthermore, the proposed iterative algorithm also improves the computational efficiency of image dehazing. Finally, the atmospheric light is predicted by the sky region in the segmentation result, and a haze-free image is obtained via the wavelength related physical imaging model. Experiments conducted on 500 synthetic images and real-world images have demonstrated that our algorithm can improve dehazing precision by at least 7% and shorten running time by roughly 15% compared to existing dehazing methods. Hence, the proposed method can be used for image dehazing in the traffic monitoring system.