Page 162 -

3.5 Pyramids and wavelets 141

(a) (b)

(c) (d)
Figure 3.41 Laplacian pyramid blending (Burt and Adelson 1983b) c 1983 ACM: (a) original image of apple,
(b) original image of orange, (c) regular splice, (d) pyramid blend.

blended, while the higher-frequency textures on each fruit are blended more quickly to avoid
“ghosting” effects when two textures are overlaid.

To create the blended image, each source image is first decomposed into its own Lapla-
cian pyramid (Figure 3.42, left and middle columns). Each band is then multiplied by a
smooth weighting function whose extent is proportional to the pyramid level. The simplest
and most general way to create these weights is to take a binary mask image (Figure 3.43c)
and to construct a Gaussian pyramid from this mask. Each Laplacian pyramid image is then
multiplied by its corresponding Gaussian mask and the sum of these two weighted pyramids
is then used to construct the final image (Figure 3.42, right column).
Figure 3.43 shows that this process can be applied to arbitrary mask images with sur-
prising results. It is also straightforward to extend the pyramid blend to an arbitrary number
of images whose pixel provenance is indicated by an integer-valued label image (see Exer-
cise 3.20). This is particularly useful in image stitching and compositing applications, where
the exposures may vary between different images, as described in Section 9.3.4.