Demo - Joint segmentation and estimation scheme in bi-distributed transparency (Back)

The demo shows the intermediate steps of the joint segmentation and estimation scheme in bi-distributed transparency, when processing a reaslistic synthetic sequence. We present
    The ground truth on the top left.
    The results from the bitransparent block registration on the top right. We present two images, the grayvalue of each one corresponds to the confidence measure of the estimates (the blue vectors on the right, the red on the left) on the blocks (that depends on the texture on the block). White is for the coefficient 1 (and thus very reliable estimate), and black for 0.
    The simplified affine motion fields extracted with the Hough transform on the bottom left (on the left only the blocks whose estimates vote in the accumulation matrix are pictured, on the right the extracted fields are attributed all over the image).
    The iterative minimization of the MRF energy on the bottom right (on the left the motion field estimation when the segmentation is set , on the right the segmentation when the motion fields are set).
The overall framework runs in about 90 seconds for 288 x 288 images.


Back