A research team at Disney in Zurich has developed a computer program that can create 3D models from ordinary photographs.
Using a specially-designed algorithm, the program analyses the light and shade areas of an image to work out the depth of certain objects. It then creates a 3D reconstruction of the scene, which can be used to create lifelike scenes for Disney films and video games, and eventually be used to make, or print, accurate models of family portraits, days out or wedding shots.
However, the program needs multiple photos taken from different angles, of the same scene, to create a full reconstruction.
HOW DOES IT WORK?
Multiple, high-resolution photos are taken from different viewpoints of the same scene. The Disney program scans the image and analyses the different light rays around objects to work out their respective depths within a scene. An algorithm assess this light ray data to build a 3D reconstruction. Once a photo has been scanned and reconstructed it is then rendered, bottom right. These rendered images could be used to create lifelike scenes for Disney films and video gamesThe research is published in a paper called Scene Reconstruction from High Spatio-Angular Resolution Light Fields.Its authors Changil Kim, Henning Zimmer, Yael Pritch, Alexander Sorkine-Hornung and Markus Gross explain that the software uses 'a method for scene reconstruction of complex, detailed
environments from 3D light fields' using light rays to capture the 'real world in unparalleled detail'.
Previous research and software has been able to capture similar details but has not been able to accurately use the data to reconstruct an accurate image.Others, including laser scanning techniques, traditionally produce a less accurate model and the laser-scanned scenes need to be touched up in post production.
Kim and his team claim that their algorithm solves this problem because it makes 'reliable depth' predictions around the outline of objects, rather than the object itself. It does this using individual rays of light. Objects that don't have a strong contrast between light and shaded areas are processed using 'fine-to-coarse' processes that looks at each object, pixel-by-pixel, in closer detail to highlight slight light fluctuations.
The authors said: 'This allows our algorithm to retain precise object contours while still ensuring smooth reconstructions in less detailed areas.'The algorithm can then process the light fields information on a standard computer.
This method makes the reconstructions 'highly detailed' and the higher quality the images, the more accurate reconstruction.?
Disney Research could use the technology to build and create accurate reconstructions of scenes for its films and video games, or the models could be printed. The paper adds: 'Scene reconstruction in the form of depth maps, 3D point clouds or meshes has become increasingly important for digitizing, visualizing, and archiving the real world, in the movie and game industry as well as in architecture, archaeology, arts, and many other areas.'