3D Images Ready to Take Leading Role

A novel technology created by School of Engineering researchers is set to change the face of films and computer games by simplifying the conversion of two-dimensional photographs into three-dimensional images. The semi-automatic, image-based computer modeling technique will enable movie-makers and games designers to easily and cost-effectively produce 3D models, slashing the time and human effort currently required by 90%, according to Prof Long Quan, Computer Science and Engineering.

Prof Quan, who has spent over 10 years researching computer vision and imaging, explained that production costs for 3D images were now high due to the expensive 3D scanners and enormous amount of work required from digital artists to reproduce high-quality imitations of real objects. For example, a 3D image of the Parthenon, the classical Greek temple in Athens, might take five people a week to capture the data on site, then two digital artists a month to reproduce it. “Using our technology, only one person would be needed for just a few hours to take digital photographs and shoot some video for the computer to process,” he said. “It would take less than 10% of the effort now spent.”

After inputting a series of 2D pictures taken from different angles, the HKUST technology identifies an object’s characteristic points, finds the corresponding points in the different images, computes the results, and is then able to reconstruct a realistic 3D image. The data is stored in VRML. If the computer collected too many or too few characteristic points it would be impossible to recreate the image, Prof Quan said. However, by utilizing the “quasi-dense” technique, which collects sparsely in some areas and densely in others, the HKUST technology can quickly and effectively transform a 2D image into 3D.

"With about 20 2D pictures, a 3D image or model can generally be reproduced in around five minutes," Prof Quan siad. "You don't have to take the 2D images in any special way either. Anyone with a digital camera can do it. Our example of a tree (pictured) which will be featured at the prestigious computer graphics conference SIGGRAPH 2007 in August shows that this is the first image-based tree modeling method that produce a quality that could be adequate for movies."

Prof Quan said that local post-production companies had already shown interest in the research. The University is working together with a media company and he hoped the technology would be transferred to industry soon.