Research Of Seabed Video Image Restoration Algorithms And Mosaic Technology

Underwater optical vision is widely used in underwater vehicle vision,platform installation and maintenance,positioning and navigation because of its rich details and high resolution.However,due to many suspended particles,serious attenuation of light wave and inconsistency of red,green and blue channels in underwater,which make the underwater image blurred seriously,low contrast,narrow field of vision and bluish-green image,all of these seriously affect the visual effect and the perception ability of underwater scene.In this paper,a set of underwater video image restoration and splicing technology solutions is proposed.The restoration improves image quality and compensates for color distortion.And the image mosaic solves the problem that underwater images can't simultaneously take wide field of view and high resolution contradiction into account.The scheme is verified by real seabed video.The results show that the proposed method can obtain underwater high-quality panoramic image with wide field of view,little color distortion,clear details,smooth transition and good visual effect from the fuzzy undersea video.This paper includes video key frame extraction,underwater image restoration,feature-based image registration,image fusion and stitching.By recovering and splicing a plurality of narrow-view image sets extracted from underwater video,a high-quality panoramic image with wide field of view is obtained.The main work is as follows:(1)This paper studies the implementation process of the basic K-means clustering key frame extraction algorithm.Experiments show that the basic algorithm is sensitive to the initial clustering center and needs to determine K value beforehand.This paper uses equal interval to set the initial clustering center to improve the random setting of the basic algorithm,but still needs to set the appropriate K value beforehand.To improve the problem,this paper proposes an automatic clustering method by accumulating inter-frame difference.The results show that the improved algorithm has a high recall rate while guaranteeing precision ratio.(2)Based on the principle of dark channel prior(DCP),the performance of conventional DCP restoration algorithm based on guided filtering and fast DCP restoration algorithm based on fast guided filtering is compared.The results show that the restored image quality obtained by the two algorithms is almost the same.Fast DCP recovers faster than conventional DCP.In order to solve the problem that DCP fails in some underwater scenes,this paper introduces non-local prior(NL)for underwater image restoration.The results show that the NL underwater restoration image is more thorough than Fast DCP underwater restoration image in fog blur removal.Restored image is clearer and brighter with more prominent details,higher image quality and better than the dark channel prior.The average gradient of Fast DCP restored image is higher than that of original image,and the NL restored image is further improved on the basis of Fast DCP restored image.NL has a good effect on images with low water degradation quality.(3)Aiming at the problem of color distortion caused by inconsistent attenuation of channel in underwater images,this paper proposes an improved underwater image restoration algorithm based on NL,which uses the measured attenuation coefficient ratio of Jerlov water type to compensate each channel,so that in the compensation space the channel has a consistent transmittance.Experiments show that the improved NL algorithm can effectively correct the underwater image color distortion when there is a matching attenuation coefficient ratio,and the restored image quality is further improved than the NL restoration image.(4)Image registration,fusion and masoic based on Harris features were studied,including detection of Harris corner points and non-maximal suppression.Descriptors were constructed by geometric fuzzy.K-d tree was used to combine with the nearest neighbor to the next nearest neighbor ratio method for rough matching,and RANSAC algorithm was used to eliminate mismatches.Experiment compares the processing effects of direct superposition unfused,average weighted fusion and progressive gradual de-weighting fusion algorithm.The strategy of image mosaic synthesis is optimized.By combining frame image with mosaic image using intermediate frame as initial reference image,the cumulative error of traditional synthesis method is overcome,and the distance from each mosaic frame to initial reference image is effectively balanced.The comprehensive experiment of underwater video image is carried out to verify the feasibility and validity of the underwater video image restoration and splicing scheme.