Research On Underwater Image Compression And Restoration Method Based On Perception Driven

Underwater optical images directly reflect the underwater situation,and play an important role in the seabed topography survey,aquaculture and other aspects.Due to the selective attenuation characteristics of light and the influence of image compression and transmission,the displayed image on the ground may appear blurry,noise and color shift,which hinders the extraction of underwater information.Image quality perception technology can replace the Human Visual System(HVS)to perceive the degradation in the image,and perform quality supervision and guidance during the underwater compression and ground-side restoration stages.Image quality perception technology helps display more effective image data on the ground.Therefore,the perception-driven underwater image compression and restoration method studied in this paper is of great significance.The specific work is as follows:(1)We analyze the typical degradation factors of underwater images from acquisition,compression to transmission.These degradation factors include color degradation,compression damage and transmission degradation.Then we introduce the design of the quality perception module and its quality supervision function for different degradation factors.The introduction in the first chapter provides a theoretical basis for the following work.(2)We propose an image compression method based on the Just Noticeable Difference(JND)model to optimize the compression of underwater images.We compare some Compression Sensing(CS)methods based on the needs of underwater tasks,and select the appropriate compression algorithm used in this paper;we propose an underwater image JND model,which includes a compressed damage perception network and a feedback-based search strategy.The experimental results show that,compared with the existing JND model,the error between the JND image predicted in this paper and the ground truth is the smallest under multiple indicators such as peak signal-to-noise ratio.In other words,the JND model we proposed can help the system remove data redundancy as much as possible without impairing the visual experience.(3)We propose a perception-driven underwater image restoration method to repair noise caused by transmission degradation and color shift caused by color degradation.We use denoising network based on fully convolutional network to repair CS transmission degradation,and design block-based transmission degradation perception algorithm as a part of the loss function.Experimental results show that compared with other partialreference perception algorithms,this perception algorithm can more accurately reflect the degree of image transmission degradation and has good error resistance performance.In addition,the performance of the denoising network combined with the perception algorithm is better than that of the original network;We design a region of interest(ROI)based color degradation perception algorithm,which is used to evaluate whether the image received on the ground needs color degradation repair and dynamically select the appropriate enhancement algorithm.Experimental results show that,compared with other no-reference methods,the correlation between the proposed color degradation perception algorithm and subjective perception is better,and it can better complete underwater tasks.