Graphics Reference
In-Depth Information
(a)
(b)
(c)
Figure 3.16.
A multi-image photomontage created with
α
-expansion. (a) The original images,
color-coded as red, green, and blue. Each image has some unsatisfactory facial expressions. (b) A
user scribbles on faces and body parts to keep from each source image, resulting in the labeling
map at right. Note that in several cases, a person's head is taken from one source image and
their body from another. (c) The final composite.
We can modify Equation (
3.27
) similarly to Equation (
3.24
) to bias seams to lie along
existing image edges.
Unfortunately, the graph-cut algorithm cannot directly minimize a function like
Equation (
3.25
) wherewe havemore than two possible labels per pixel (i.e., more than
