Due to the absorption of seawater and the scattering of suspended particles in water, the deep-sea image obtained by underwater robot through artificial light source is generally fuzzy, color deviation, and low resolution. Focusing on the key problems to be solved in the rapid and accurate restoration of deep-sea images, the data set of real deep-sea images is firstly established, and the imaging characteristics of deep-sea images are analyzed. Based on the statistical results of image features, a linear depth of field model is proposed. Then, the model parameters are identified by supervised method. Finally, according to the depth of field model, the transmission map and background light of the original image are estimated quickly, so as to effectively avoid cumulative error and achieve effective restoration of deep-sea images. Experimental results show that the proposed algorithm is superior to other algorithms in terms of image restoration results, validity, quality, and real-time performance. Processing 600 pixel×800 pixel image on Nvidia Jetson TX2 embedded device, the average restoration speed of the proposed algorithm is 3.08 times faster than the four outstanding underwater image enhancement algorithms.