Skip to content

Simulating viewpoint changes on a stereo image pair

by werner on January 7th, 2010

The Fujifilm Finepix Real 3D W1 has a 3D screen which allows to perceive the depth of a stereo image without having to wear special glasses. Now, how can we convey depth on a regular computer screen?

I took a take on this, by creating a video sequence which transitions between the left and the right ‘eye’ of the two stereo images. Click on the image below to see the video.

A simulated viewpoint transition

A simulated viewpoint transition

How does this work?
The videos make use of disparity maps between a stereo image pair. A disparity map describes the distance between corresponding points on two images.

The video gradually displaces the pixels from their original location up to the distance given by the disparity maps. Pixels on the left image which become obscured by the transition are filled in by pixels taken from a transition computed from the right image.

I took the algorithm for computing the disparity maps from the publication «Depth Discontinuities by Pixel-to-Pixel Stereo» by Stan Birchfield and Carlo Tomasi.
The algorithm works with gray scale images. To get good results with color images, my program converts them into the YCC color space, and then uses the Y-channel as input for the algorithm.

If you want to experiment with this algorithm by yourself, I recommend downloading the C source code from the publication.

If you are not a programmer, you can play around with my 3D Image Laboratory software.

From → Image & Video, Java

Comments are closed.