Due to the light absorption and scattering effect of the water volume, underwater images suffer from strong attenuation, high noise and serious color distortion. Thus, the original underwater images can hardly meet the requirements of visual observation and image analysis. To solve this problem, this article proposes an underwater image restoration method based on gradient prior for improving underwater image quality. First, according to the underwater imaging prosperity a gradient prior is established for underwater images that the gradient of scene imaging information (the clear layer of underwater images) must be larger than that of scattering information ( the haze layer of underwater images). Secondly, the gradient prior is modeled with the underwater imaging model. Finally, a semi-quadratic optimization function is utilized to extract the object reflection information which generates final image restoration results. Underwater samples in UEIB and RUIE datasets are taken for experiment evaluation. In contrast to five state-of-the-art underwater image information processing methods, the proposed method obtained prominent performance in both qualitative and quantitative evaluation. The average PSNR, SSIM, and UIQM values of this method are 20. 066 5, 0. 696 1, and 3. 902 9, which are better than compared methods. Therefore, the proposed method can efficiently remove underwater image noise. The image signal to noise ratio), sharpness and the contrast of underwater images are enhanced.