Research On Defogging Methods Of Dark Primary Color Image Based On Visual Information Fidelity

Image is one of the most commonly used information carrier in the information-based society,the quality of image directly affects the realization of subsequent computer vision applications.In abnormal weather conditions(such as fog,rain,haze,etc.),images captured often suffer from reduced contrast and distortion.Since the distance from the imaging system to the target object and the attenuation caused by the atmospheric microparticle are unclear,image dehazing becomes a challenging task.The proposed haze degradation model provides a new idea for haze removal,with the aid of prior information,the defogging efficiency of the model is greatly improved.Among these priors,the most convincing one is the dark primary color prior theory,but it also has disadvantages such as insufficient brightness of restored images and halo effect.Therefore,in this study,two fidelity dehazing algorithms,the brightness fidelity dehazing algorithm based on the HSI color space dark primary color and the detail information fidelity dehazing algorithm based on the region segmentation dark primary color,were investigated according to the dark primary color theory.(1)Research on brightness fidelity haze removal algorithm based on dark primary color prior of HSI color space.The original dark primary color model was transformed from RGB color space to HSI color space,pixel color concentration and brightness difference features were introduced to raise the transmission from patch-level to pixel-level,the transmission was refined by guided filtering.Subsequently,in view of the problems of low saturation,high brightness and fuzzy details in the dense fog region,these parameters were normalized and weights were constructed to recalculate the atmosphere light.Finally,image restoration was carried out based on foggy degraded images.The local transmission similarity of traditional patch prior strategy was preserved by image block brightness,and pixel feature differences were introduced through color concentration and detail,which protected the saturation and detail.Experiments with the comparison algorithm verified that the improved algorithm recovery results were closer to the original scene and agree with human vision perception.(2)Study on detail information fidelity and haze removal algorithm based on region segmentation dark primary color.By improving the minimum filter of the dark primary color prior algorithm,the edge information of the image after defogging was preserved effectively.The improved filter divides the image into high brightness and low brightness regions.Minimum filtering was still used in low-brightness region,and linear attenuation was used in high-brightness region to preserve the original features.The transmission was calculated by combining dark primary color prior and haze degradation model,followed by minimum filtering and guided filtering were performed to optimize the transmission map.In addition,the introduction of closed operations has improved the accuracy of atmospheric light intensity,and then improved the haze degradation model to protect the image brightness after haze removal.After comparison experiments,we found that the algorithm in this study has better performance in protecting detail edge information and suppressing halo effect.