CHAPTER 11 HOW THE DIGITAL DARKROOM MAKES GOOD PHOTOS INTO GREAT FANTASIES^179
The program checks the images with the greatest number of feature matches for an image match. A RANSAC
(RANdom SAmple Consensus) counts a random number of matching features that are within a certain distance of
their statistically predicted locations. The sample set with the most features more or less where they should be,
calledinliers, is kept as the working solution.3
Using the prevailing set of matching inliers, Autostitch determines the geometrical changes between images. This tells the program
which pixels on the edge of one photo should be joined to pixels along the border of another photo. Doing so is likely to require dis-
torting, twisting, and resizing of the pictures to overcome incongruities caused by perspective, random camera movements, and the
characteristics of the camera’s lens. Autostitch projects the stitched images onto a flat plane, but other programs, such as those that
produce the 360-degree panoramas of family rooms and kitchens seen on real estate websites, wrap their stitchwork onto cylindrical
or spherical surfaces.4
The regions where the pictures join are likely to have
inconsistencies caused by different exposures, parallax
errors, or people, clouds, or traffic present in one image
but not another, producing ghosts, semitransparent or
truncated objects. The ghost problem is solved by elimi-
nating the person or other object creating the apparition.5
Autostitch resolves differences in hue and brightness through multi-band blending. The program samples several pixels
from a broad stripe at the photos’ intersection, averages them, and repeats the process until the change is made gradually—
and hopefully unnoticeably—from one image to the other.6