Research On The Technology Of Improving Clarity In Underwater Images

With the expansion of the earth's population,the shortage of land resources and the deterioration of ecological environment,human beings have gradually focused their attention on the ocean,which accounts for 71% of the earth's surface area.Underwater images frequently play a key role in the fields of marine biology research,seabed resource exploration,marine military applications,and underwater archaeology.Especially,the underwater images contain richer details and color information than acoustic images.Thus,they have an irreplaceable status in underwater scientific research and engineering practice.Moreover,high-quality underwater images can not only meet people's needs of visual perception,but also contribute to the application of high-level vision.However,due to the uniqueness and complexity of underwater imaging environment,the underwater images mostly have varying degrees of degradation.For example,light absorption causes the color shift and low illuminance of underwater images,and light scattering causes the blur of underwater images.Multiple degradation in underwater images reduces the clarity,and seriously affects the visual perception and application requirements of images.Therefore,based on the characteristics of light attenuation in water,the technology of images enhancement is firstly used to improve the clarity of underwater images.Then,the neural network model based on the deep learning is utilized to enhance the clarity of underwater images.(1)Aiming at the problems of color shift,foggy blur,low exposure and non-uniform illumination in underwater images,an underwater images enhancement algorithm based on color attenuation compensation and Retinex is proposed.Firstly,in order to correct the color shift of underwater images,the R,G,and B channels of underwater images are adaptively compensated by utilizing water's inconsistent attenuation characteristics to different wavelengths of light.Then,Retinex based on multi-scale guided filter is used to remove the foggy blur and enhance the contrast.Finally,underwater images are normalized according to the features of histogram distribution in underwater images and natural images,so as to enhance their texture and exposure of images on the premise of preserving the main information.Compared to other algorithms,the experimental results show that the proposed algorithm not only has the better visual perception,but also has the higher evaluation score of image quality.The proposed algorithm has the characteristic of strong adaptability,which is helpful for the application of computer vision algorithms in the water.(2)Due to the phenomenon and degree of degradation in underwater images are characterized by diversity,it is difficult for traditional image processing technology to extract universal prior knowledge.Therefore,a new underwater images enhancement algorithm based on color separation and residual learning is proposed.Firstly,considering the attenuation difference of channels in underwater degraded images,the brightness channel and the chroma channel are creatively separated.Then,a convolutional neural network model for underwater images enhancement is designed based on the codec framework and residual learning,and the designed neural network is utilized to train the separated channels.Finally,the outputs of network are post-processed to obtain highdefinition underwater images.Experimental results show that the proposed algorithm can significantly improves the clarity of underwater degraded images.The enhanced underwater images have higher quality of visual perception,and are more in line with the subjective perception of human eyes.