To enrich 3D scenes, a real–virtual fusion-based integral imaging method is proposed. It combines the Softargmax function with Gaussian weighting coefficients for sub-pixel feature point extraction from SuperPoint detection results. SIFT is also used for feature point detection and matching, along with the improved SuperPoint. Subsequently, based on the multi-view 3D reconstruction, the real object is reconstructed into a 3D model. A virtual model is then fused with the 3D reconstructed model of the real object to generate a real–virtual fusion elemental image array based on the display platform’s optical parameters. The experimental results demonstrate that the proposed method can optically reconstruct more realistic and vivid real–virtual fusion 3D images. This method can enrich a scene’s content, enhance visualization and interactivity, save costs and time, and provide flexibility and customization.