Robot Vision Group

National Laboratory of Pattern Recognition

Institute of Automation, Chinese Academy of Sciences

2D to 3D Conversion for Films

(Supproted by Natural Science Foundation of China)


Nowadays, 3D films are becoming more and more popular. Because 3D images could provide higher realism than the conventional two-dimensional (2D) images, cinemas show an increasing number of 3D films, companies are also beginning to launch 3D mobile phones and offering 3DTV sets. However the content created directly in some suitable 3D format is difficulty: an expensive stereo-camera system is needed for capturing 3D images, and when taking images, highly-skilled camera technicians are required, who can control the camera vergence appropriately. Since there exists a tremendous amount of media data in 2D format, a high demand for adding 3D effect to them is growing. This is where the 2D to 3D conversion method comes to rescue.

What is a 3D Film?

A 3D (three-dimensional) film or S3D (stereoscopic 3D) film is a motion picture that enhances the illusion of depth perception. Derived from stereoscopic photography, a regular motion picture camera system is used to record the images as seen from two perspectives (or computer-generated imagery generates the two perspectives in post-production), and special projection hardware and/or eyewear are used to provide the illusion of depth when viewing the film. 3D films are not limited to feature film theatrical releases; television broadcasts and direct-to-video films have also incorporated similar methods, primarily for marketing purposes.

Fig.1. Principle of stereoscopic vision

The principle of 3D films is based on stereoscopic vision. It is a simple system of stereoscopic vision in Fig.1. Two cameras with the same parameters are placed in positions with distance B apart along coordinate x (When making films with stereoscopic cameras, B is set to 64mm which is the average distance between the two eye pupils of an adult). Additionally, their axes parallel to each other in the same direction with coordinate z. Image planes I1, I2 are also parallel to coordinate plane (x, y). O1, O2 are focus and Pl, Pr the images of 3D point P for the two cameras respectively. f is the camera focal length. Thus parallax is defined as: D=Pl-Pr.

Some Results

According to the relationship between movements of camera and scenes, there are four cases as follows:

  1. Camera moves and at least one object moves in the scene.
  2. Camera does not move while at least one object moves in the scene.
  3. Camera does not move and no object moves in the scene.
  4. Camera moves while no object moves in the scene.

Now the system we have developed focuses on the fourth case. We proposed an approach of 2D to 3D conversion based on structure from motion. Some examples of synthesized anaglyph images of the evaluation test sets as follows. (Click the image to see large version. Note: a red and cyan 3D glass is needed)



Fig. 2 Some synthesized anaglyph images. The 3D effect can be seen with a red and cyan 3D glass