At present, projection compensation algorithms have achieved good research results, but most of the projected image color compensation research ignores the optical part of the color transfer function modeling process, resulting in poor modeling accuracy of the color transfer function. At the same time, most of the deep learning network optimization designs are less for the phenomenon of deepening the network resulting in the loss of extracted feature information in the process of projected image colour compensation. To address the above problems, a luminosity compensation method for projected images based on attentional feature enhancement is proposed in this paper. The method extracts feature information from the projected surfaces with colored textures by increasing the depth of the network, and employs deep learning to fit a complex composite radiative transfer function to solve the problems of traditional photometric compensation methods, which improves the quality and colors of the projected images, and further eliminates the reliance on high-quality projection screens. The luminosity compensation results of the proposed method in this paper are better than other comparative algorithms in three evaluation indexes, Peak Signal-to-Noise Ratio (PSNR), Root Mean Square Error (RMSE) and Structural Similarity Index Measure (SSIM). Compared with the CompenNet series of methods, the proposed method in this paper improves up to 5.717% in PSNR evaluation metrics, reduces up to 14.968% in RMSE evaluation metrics, and improves up to 2.893% in SSIM evaluation metrics.