A Single Traffic Image Fast Dehazing Method with Sky Segmentation #br# and Local Transmittance Optimization
LI Xi-ying1,2,3, ZHU Ken-gang1,2,3
2019, 0(05):
51.
doi:10.3969/j.issn.1006-2475.2019.05.010
Asbtract
(
176 )
PDF (6398KB)
(
145
)
References |
Related Articles |
Metrics
Traffic photographs taken in hazy weather are degraded, due to the suspended particles in the air and scatter light. Since the scattered environment light is mixed into the light accepted by the observer, the contrast and sharpness of the hazy traffic image decrease, and the difficulties of subsequent processing and analysis increase. These problems, directly influence the full play of the circuit television surveillance system utility. Therefore, fast and effective traffic image dehazing has important application value. In the existing dehazing algorithm, the transmittance estimation deviates quite greatly from the actual situation. Especially when dealing with the sky area, it is easily lead to problems such as color distortion and halo effect. On the basis of the dark channel prior theory, this paper puts forward a fast haze removal method integrating sky segmentation with local transmittance optimization. First, the original traffic image is segmented into sky area and non-sky area by OTSU. Secondly, on the basis of the dark channel prior, the transmittance of the non-sky area is optimized by the maximum filtering and guided filtering, and the transmittance of the sky area is corrected by adaptive parameter adjustment method. In the end, the restored image is adjusted by Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve the image brightness. Experimental results show that the proposed algorithm can effectively reduce the phenomenon of color distortion and halo effect in the sky area, and obtain a more natural and clear restored result. For the non-sky area, the clarity and contrast of the restored result are higher, and besides, the proposed algorithm keeps high efficiency. What’s more, compared with the dark channel prior algorithm, Tarel algorithm, Meng algorithm, Zhu algorithm and Berman algorithm, the proposed algorithm does better in terms of variance, average gradient, image information entropy and other indicators. This proposed algorithm can effectively and quickly restore the haze traffic image and reduce the color distortion and halo effect in the sky area. The restored image has good clarity and color revivification degree, and obtains better image sharpness and contrast enhancement. The proposed algorithm can provide good theoretical and technical support for the road traffic supervision.
