Research On Defogging Algorithm Of Physical Model Image Based On Sky Area Recognition

In the haze weather,there are suspended particles with large plasma radius,for example dust and water droplets.The light of the target scene is affected by these suspended particles during the transmission process,such as refraction,scattering,absorption and so on.Therefore,the images observed in the imaging equipment and in our human visual system are different from the actual scene images on sunny days.The contrast of foggy images is decreased and the degradation is serious.It has an impact on the use of various aspects of the visual system,such as travel traffic,ship travel,aircraft aerial photography,road monitoring and so on.Therefore,it is very important to get high contrast,rich details and clear and accurate fog-free images.In this paper,the physical model and dark channel de-fogging algorithm are used to remove the fog in image.The main contents include two parts: the thinning of non-sky region and the recognition and segmentation of sky region.In this paper,firstly,aiming at the problem that dark channel prior algorithm is too rough in estimating transmittance,it will produce block effect,and finally make the restoration result appear halo phenomenon,a method of thinning transmittance is put forward.According to the dark channel algorithm,the transmittance to be thinned is obtained first,and then the modified fine transmittance is obtained by combining the details of the original image after sampling.For the calculation of atmospheric light value,the quaternion method with small prediction error is adopted,and finally the removal image is obtained.The dark channel prior algorithm appears halo phenomenon when dealing with the sky region,so in this paper,we first identify whether the image contains the sky region or not according to the characteristics of the sky region.After the sky region is identified,the sky region can be segmented by using the characteristics of large brightness and low gradient change.In this paper,we compensate for the information in the sky region,set the appropriate threshold to increase the transmittance and avoid the color deviation in the sky region.For the non sky region,the refined transmittance method is used to combine theatmospheric physics model to get the fog image.In order to further reduce the error,the pixels in the abrupt edge region are processed and fused to get the final fogging effect.In order to verify the validity of the final results,this paper mainly through two aspects to verify.Firstly,the subjective evaluation indexes observed by the observer directly are selected randomly;secondly,the objective data evaluation indexes such as peak signal-to-noise ratio,information entropy,average gradient and structural similarity are calculated by calculating the image's peak signal-to-noise ratio(PSNR),information entropy,average gradient and structural similarity.Finally,the running time of several algorithms is analyzed,and it is concluded that the processing time of this paper is better than that of dark channel algorithm.The experimental results show that compared with the dark channel algorithm,the proposed algorithm has a good effect on the image recognition and fog removal from the sky region.