Abstract: In order to achieve  more high-quality images in 3D digital image virtual scene  reconstruction, the author proposed  a 3D digital image virtual scene reconstruction algorithm based on machine learning. Firstly, the state information and presentation instructions of the scene were analyzed to obtain the vertex distribution positions of the mesh model for image reconstruction. The global image was approximately calculated by local coordinate method, and the local details were corrected to complete the rendering processing of 3D digital images. Secondly, taking space and scale as feature points, a window detection template was constructed on the image, and a classifier was used to suppress discrete feature points and remove redundant features. Finally, according to the fitting function method, the smoothed 3D coordinates were obtained  to reconstruct the 3D surface, and the local 2D triangle was segmented and mapped to the 3D space to realize the 3D digital image virtual scene reconstruction. The experimental results show that the proposed algorithm converges faster, the reconstructed image details and edge contours are complete, and the overall effect is good.

Key words: three-dimensional digital image, machine learning, virtual scene reconstruction, digital image processing, feature extraction