This paper proposes an image gradient extraction method that fuses luminance information with chromaticity information. The method introduces the CIE-L*a*b* color model based on the human eye vision model to obtain the chromaticity gradient information on the basis of retaining the luminance gradient information, while the normalized luminance gradient and chromaticity gradient are later fused with the gradient. Afterwards, to balance the local contrast and color fidelity image enhancement effects, the Retinex algorithm with multiple scales was selected to better process the image. An end-to-end small-scale CNN model is constructed for color correction, and then a multi-scale Retinex model is used for texture enhancement, which integrates the color restoration and texture enhancement of the image, and the texture blurring and color deviation problems of the underwater image. The results show that the entropy value of the image increases from the original 6.5847 to 7.6014 after the processing of the method in this paper, which shows that the entropy value of the image can reflect the difference of the image contrast. After qualitative and quantitative analysis, it is found that the color bias of the underwater image enhanced by the method of this paper is effectively corrected, and the problem of color distortion is significantly improved. Meanwhile, the contrast, sharpness and saturation of the image are improved significantly, which proves the superiority of the proposed method in this paper compared with other methods in underwater image enhancement.
