Page 214 -

Section 6.4 Image Denoising 182

Initial Image Object masked out

Initial Image Object masked out Object composited back

Initial Image Hole Extended by hole filling

FIGURE 6.15: Modern hole-filling methods get very good results using a combination of
texture synthesis, coherence, and smoothing. Notice the complex, long-scale structure
in the background texture for the example on the top row. The center row shows an
example where a subject was removed from the image and replaced in a different place.
Finally, the bottom row shows the use of hole-filling to resize an image. The white block
in the center mask image is the “hole” (i.e., unknown pixels whose values are required to
resize the image). This block is filled with a plausible texture. This figure was originally
published as Figures 9 and 15 of “A Comprehensive Framework for Image Inpainting,”
by A. Bugeau, M. Bertalm´ıo, V. Caselles, and G. Sapiro, Proc. IEEE Transactions on
Image Processing, 2010 c IEEE, 2010.

6.4 IMAGE DENOISING
This section addresses the problem of reconstructing an image given the noisy ob-
servations gathered by a digital camera sensor. Today, with advances in sensor
design, the signal is relatively clean for digital SLRs at low sensitivities, but it
remains noisy for consumer-grade and mobile-phone cameras at high sensitivities
(low-light and/or high-speed conditions). Adding to the demands of consumer and
professional photography those of astronomy, biology, and medical imaging, it is
thus clear that image restoration is still of acute and in fact growing importance.
Working with noisy images recorded by digital cameras is difficult because different
devices produce different kinds of noise, and introduce different types of artifacts
and spatial correlations in the noise as a result of internal post-processing (demo-
saicking, white balance, etc.).