| The rapid development of artificial intelligence technology has also led to a gradual increase in the fields it involves.Among them,the field of computer vision through the simulation of biological vision to achieve the relevant intelligent devices to image acquisition and information conversion,so that artificial intelligence technology in aerospace photography,traffic video screen monitoring,terrain survey,medical and other aspects to gain wide attention and application.Therefore,this research area of computer vision has become a popular direction in artificial intelligence,but its related image or video detection equipment is highly susceptible to bad weather,such as fog,haze and other bad weather,when taking images outdoors.This is because the formation of fog and haze is due to the atmosphere contains a lot of suspended media resulting in light scattering or absorption,making the human eye visual visibility reduced,such as: water droplets,aerosols,sand and dust particles.This weather phenomenon can make some images or video capture devices present problems such as low contrast,color shift,and loss of details,resulting in difficulties in extracting and using image information features,limiting the application of various outdoor monitoring and detection systems,and causing serious impacts on tasks related to the field of computer vision.Improving the quality of hazy images and increasing the robustness of the system under inclement weather conditions is thus of great scientific importance and wide application value.Considering that the atmospheric scattering model describes in detail the degradation process of images captured in hazy scenes,this paper proposes three different dehazing algorithms using the atmospheric scattering model as a constraint,combining the prior knowledge with hypothetical conditions and analyzing the advantages and shortcomings of existing algorithms,The details of the study are as follows:(1)To address the problems in image dehazing such as inaccurate scene depth and atmospheric light estimation,which lead to the degradation of image quality acquired by imaging devices,a dehazing algorithm combining linear scene depth estimation and adaptive hazy concentration estimation is proposed.First,according to the relationship between scene depth and luminance components and saturation,a linear model is established to estimate the scene depth by using double filtering to optimize the high brightness region of both,combined with linear transformation.Then,the texture features are extracted to construct a hazy concentration model to obtain the adaptive scattering coefficient,which is used to obtain the transmission.Finally,two different atmospheric light estimation methods are used according to the verdict of whether the hazy map contains the sky region or not.Through both subjective and objective analysis,the experimental results show that the proposed method has good efficiency and robustness in preserving the depth edges,color quality,and detail,and the quality of the image recovery is relatively good.(2)To address the problems of color bias and visual bias in recovery results due to inaccurate estimation of transmission and atmospheric light in the dark channel prior,an adaptive transmission and atmospheric light correction dehazing algorithm based on multiscale morphological reconstruction is proposed.First,the algorithm uses the open operation after morphological reconstruction instead of the minimum filtering operation in the dark channel,and uses the image morphological edges to set the scale of the open operation structure elements to construct a multi-scale open operation fusion dark channel,to obtain the initial transmission.For the problem of the failure of the dark channel of the sky-containing hazy image,an adaptive transmission correction model is fitted with a Gaussian function to correct the initial transmittance of the sky-containing hazy image.Then,the local atmospheric light is improved based on the image brightness information with morphological closed operations.Finally,The proposed algorithm is combined with the atmospheric diffusion model to obtain an accurate haze free image.Experimental results show that the algorithm is suitable for the recovery of blurred images in a variety of scenes,and the effect of recovery is good.(3)When the haze degradation scene is under the condition of uneven illumination,some of the scene details will not only be less visible due to hazy obscuration,but also make some regions invisible due to light shading.To address this problem,a hazy sky image restoration algorithm with double light curtain boundary constraints based on an improved atmospheric scattering model is proposed.Firstly,to avoid the influence of information such as texture noise in the image on the recovery results,the hazy image is divided into two parts: texture layer and haze layer.And the imaging principle of traditional atmospheric scattering model is analyzed,the model is improved by using its degradation mechanism combined with Retinex theory.Secondly,we introduce the mean inequality relationship and Gaussian decay function to estimate the atmospheric light curtain by estimating the eigenvalues,and set the upper and lower boundaries to constrain it.Then use the improved atmospheric scattering model to find the incident light of the scene,and use the bright channel prior to find the bright channel with hazy image to compensate for the incident light from the scene.Finally,the local atmospheric light acquisition method is improved,and the local atmospheric light estimation method based on the middle channel is proposed,and the haze layer is dehazed by combining the requested atmospheric light curtain with the scene incident light substitution into the improved atmospheric scattering model,and the resulting fog-free image is combined with the image texture layer to obtain the final recovery results.Analysis of the experimental results shows that the proposed algorithm can efficiently recover hazy images in scenes with uneven illumination or other scenes. |
PDF Full Text Request