Research On Image Defogging Algorithm Of UAV In Fog

Traditional satellite remote sensing and aerial photography are limited in application due to their high cost,low flexibility and complex technology.In recent years,UAV aerial photography has been widely used because of its flexibility,low cost and good economy.At the same time,because the fog and haze extreme weather will lead to the serious degradation of image quality collected by UAV and other aerial tools,which makes the fog image color gray,target fuzzy and image contrast reduced,which limits the application of image,and has great interference to the advanced processing of image later stage,such as feature extraction,target recognition,image fusion and so on.Therefore,defogging reduction is of great application value.Since image defogging itself is a multi-disciplinary fusion cross topic,it can not only enhance the visual defogging effect by improving image contrast and highlighting useful information,but also restore the defogging by using the physical model of fog imaging based on the principle of fog image degradation.Therefore,this paper reviews and analyzes two kinds of mainstream defogging algorithms,image enhancement and image restoration,focusing on Retinex and dark channel defogging algorithm,and analyzes and improves its principle.The specific research contents are as follows:(1)In this paper,the core idea of enhanced defog and restoration defog algorithm is discussed in detail,and the representative SSR,MSR and dark channel prior defog algorithm are studied in depth.The principle of the steps is analyzed in detail,and the shortcomings and shortcomings of the algorithm are studied in depth.(2)Based on the analysis of the causes of halo,detail loss,color distortion and poor effect of visual demisting after demisting with multi-scale Retinex algorithm,an improved MSR demisting algorithm with two color spaces is proposed in this paper.Firstly,the accurate incident component is calculated by using the double side encircling of space proximity and gray similarity to avoid the alternation of light and dark and the sudden change of depth of field For the interference of the emission component,the color distortion caused by the destruction of the color ratio of the MSR sub channel,using the feature that the brightness and the hue in the HSV space do not interfere with each other,this paper carries out the bilateral Retinex processing on the V component,stretches the saturation adaptively,and converts it back to RGB to obtain the color preserved image.At the same time,the original image in RGB space is processed by small-scale Retinex to get the image with improved details.Finally,the image processed by two color space is weighted and fused to get a defog image with good details and color.The normal truncation stretch is used to avoid the imbalance of gray scale while the reflection component is recovered and stretched.The experimental results show that the improved algorithm has good fidelity of image detail and color,and the effect of visual demisting is obvious.(3)This paper focuses on the research of the restored defog algorithm based on the dark channel.Aiming at the existing problems,an improved defog algorithm based on the sky segmentation is proposed.Firstly,the accurate atmospheric light value is obtained through the improved quadtree algorithm to eliminate the interference of the bright area on the atmospheric light value.Secondly,the dark channel is not suitable for the sky area.At the same time,the local square is assumed when calculating the transmittance of the dark channel The same area transmission will cause the transmission deviation.So this paper proposes an adaptive sky segmentation algorithm based on edge detection,which combines the Soble edge detection operator and Otsu to segment adaptively,and finally gets the fine sky segmentation result by morphological close operation to eliminate small area.Then,the transmittance of the sky region is compensated by adaptive proportion,and the non sky region uses SLIC dark channel to get the defog image.Experimental results show that the image details are prominent,the sky color distortion is improved,and the visual defogging is improved significantly.(4)The main defog algorithms such as MSR,CLAHE,ACE and dark channel are programmed and implemented.The results of the above algorithms and the two improved algorithms are compared.The subjective and objective evaluation proves the effectiveness of the algorithm.At last,the experiment of defogging and splicing is carried out for several UAV images,and the steps and methods of image defogging and splicing are discussed.