关键词: 水下图像, 颜色补偿, 卷积神经网络, 去雾, 直方图均衡化, 色彩恢复

Abstract: Underwater vision measurement has important applications in marine surveying, underwater engineering surveying, underwater archaeology and underwater environmental monitoring. However, underwater images suffer from color distortion, image blurring and low contrast, which limits the application of underwater visual measurement technology in practical environments. A color restoration method for underwater images based on color compensation and convolutional neural network (CNN) defogging model is proposed in this paper, in which the image enhancement is carried out step-by-step.Firstly,the color deviation of underwater images is analyzed, and then an adaptive color compensation strategy combined with the grayscale world white balance algorithm is used to correct underwater image color. Secondly, a CNN based dehazing model was designed to achieve dehazing processing of underwater images. Finally, the adaptive histogram equalization CLAHE method is used to enhance the contrast of underwater images. In order to prove the applicability and superiority of the proposed method, two image datasets are combined to study, and several known underwater image enhancement and restoration methods are compared. The proposed method and several compared methods are evaluated in two aspects of subjective visual effect and quantitative evaluation index. The comparison results show that compared with other enhancement algorithms, the proposed method successfully improves the clarity of the image and reduces the color deviation of the damaged underwater image when processing underwater images in various environments and has superior image color recovery compared with existing enhancement methods.

Key words: underwater image, color compensation, convolutional neural network, defogging, histogram equalization, color restoration