Research On Image Dehazing Based On Sky Segmentation And Super-pixel Optimization

Computer imaging systems are widely used in image detection,unmanned driving,military reconnaissance,and other fields,but fog,haze,and other weather conditions have a significant impact on the outdoor images they collect.This causes optical phenomena such as refraction and scattering in the process of light propagation,which causes all kinds of image quality degradation problems in the original images collected by computer imaging system,such as image visibility degradation,image color shift,and image content details missing,etc.,affecting and limiting the real image information obtained by imaging system.As a result,it is critical to clearly restore the collected image materials.Various dehazing algorithms are explored and analyzed in this paper,and two different image dehazing algorithms are proposed,with the foggy image with sky area serving as the main research object.The following are the specific research contents:(1)A single-image dehazing method combining Otsu’s algorithm with Canny operator edge detection for fused sky region segmentation is proposed is proposed to solve the problems of halo phenomenon in the sky area and fog remaining phenomenon in the traditional dark channel prior algorithm.To begin,the algorithm divides the original outdoor image into two parts: sky area and non-sky area.Second,the transmittance of the two areas is fine-tuned.For the sky area,a linear exponential function is built based on the dark channel prior algorithm’s sky transmittance,and the real sky area’s transmittance is corrected.Fast guided filtering is used to optimize the dark channel prior non-sky transmittance for the non-sky area,and the obtained regional transmittance results are superimposed to obtain the final transmittance.Finally,an improved quadtree search algorithm is proposed for calculating atmospheric light value,which is combined with mean value optimization to improve parameter accuracy.(2)To address the dehazing algorithm’s inaccurate transmittance and color distortion,a combination of sky region segmentation and super pixel segmentation is proposed for image dehazing.To begin,based on the gradient information in the original image,the first rough segmentation step is performed using the gradient energy function,and the result is refined by combining the Canny algorithm to achieve accurate segmentation of the foggy sky region.Second,for transmittance,the bright channel prior algorithm is used to calculate the partial transmittance of the sky area,the super-pixel segmentation algorithm is used to cluster the nonsky area,and the non-sky area’s super-pixel dark channel value is selected to realize the optimal calculation of the non-sky area’s partial transmittance,and the final fine transmittance value is obtained by combining and modifying the two partial transmittance values.Finally,the atmospheric light value is optimized using the differential equation,and the fog-free clear image is restored using the atmospheric scattering model.The results of the experiments show that this method has a strong ability to restore the sky area,and the estimation of atmospheric light value and transmittance is more accurate.The restored fog-free image has excellent brightness and contrast,the color is richer,and the image details are more visible.The experimental results show that this method is accurate and efficient in segmenting the sky area,with obvious details recovery and improvement in image dehazing effect.