Research On Underwater Image Enhancement Algorithm Under Low Light Condition

Illumination is the key factor that determines the quality of underwater images.Because the underwater environment is affected by factors such as water quality,water depth,and object occlusion,the lighting will be seriously insufficient,making the acquired underwater images have color distortion,low brightness and contrast,and poor sharpness.These problems bring great difficulties to the practical application of underwater images.This thesis combines image enhancement and deep learning theory to conduct research on these issues.The main work is as follows:A low-light underwater image enhancement algorithm based on white balance and relative total variation is proposed.According to the characteristic of selective attenuation of light in water,first the global illumination compensation is performed on the underwater image to improve brightness of the image.Then the white balance algorithm is used to color correction to make the color more balanced Finally,a relative total variation model is constructed based on guided filter to constrain the illumination map of the image to improve the accuracy of illumination map,thereby achieving underwater image enhancement.This algorithm can remove the color deviation of low-light underwater images,enhance the detail information,and improve the overall visual effect of the image.A low-light underwater image enhancement network model based on multi-scale dense block is proposed.Firstly,the low-light underwater image samples are synthesized based on the underwater imaging model;then the samples are used to train the network model,which uses multi-scale dense blocks as the basic unit,and uses skip connection and feature concat to extract diverse and abstract image features;Finally,the training process of the network is constrained by constructing a joint objective function of mean square error and structural similarity,and the trained network performs end-to-end enhancement on low-light underwater images.The algorithm not only can correct color of underwater images,improve brightness and contrast,but also has the advantage of fast network convergence.