This image of the Leaning Tower of Pisa in Pisa is composed of 84 images from a video sequence. We first removed the interlacing from the images and then registered the images to be composed together. The image on the front page has been rotated, cropped, and shrunk compared to this output from our program. Notice, that the rotation is consistent with the first frame of the video. Benedict was holding the camera crooked.

The first image of the video shows the reference frame used throughout the registration process.

This is the output of the processing of this image. Black square areas are where the error rate was above the cutoff, those spots are filled in with portions from other images in the composite.

The 32nd image from the sequence. Notice the interlacing effects.

After processing the 32nd image, most of the interlacing effects are removed.

The 75th image with interlacing effects.

After processing the 75th image.

Here is a divx compressed version of the original DV sequence. http://www.cs.princeton.edu/~bjbrown/mosaic/videos/pisa.avi