Riaz Hussain Junejo; Muhammad Abdul Rehman Choudhary MEE10:25, pp. 68. ING/School of Engineering, 2010.
In this thesis, a novel algorithm is invented to detect and extract the foreground from background of stereo pair of images. In this way, an attempt can be made, using this technique towards data compression of 3D TV. This automatic extraction is achieved using motion detection and region segmentation techniques. The problems in this process were identification of foreground from its background, performing automatic segmentation without user interaction, distance of camera from the foreground, different foreground objects in same plane and number of objects in stereo images.
In the proposed technique, pyramidal Lucas Kande technique is used to detect the most noticeable motion in the pairs of stereo images. The object with such type of motion will be detected as foreground. Its boundary points will be detected and provided to Grab Cut algorithm for effective extraction of foreground. Results of this approach will confirm our successful attempts towards data compression for the new world of 3D TV.