Digital Signal Processing Reference
In-Depth Information
4.2
The Real Image Inpainting
Figure 5 is the inpainting result of lady, the sub-image is enlargement image of local
region. From figure 5 (c), the result is close to what one would expect, although the
final image had a little indentation. From the sub-image of figure 5 (b), at the palm,
some edges are not inpainting completely and a bit fuzzy, but in sub-image of figure 5
(c), the connection of edge is more natural.
(a) (b) (c)
Fig. 5.
Lady Image inpainting result: (a) damaged image and local enlargement image (b) result
with original TV algorithm and local enlargement image (c) result with new algorithm and local
enlargement image
(a) (b) (c)
Fig. 6.
Natural color image inpainting result: (a) damaged image and local enlargement image
(b) result with original TV algorithm and local enlargement image (c) result with new algorithm
and local enlargement image
In figure 6, we display the result for temple color image, where we applied our
method to three channels simultaneously. Relatively speaking, the final result is good
for this numerical scheme of our algorithm.
By both subjective observation and mean square error evaluations with respect to
the originals, we conclude that the proposed method gives satisfactory output for
structural inpainting. Table 1summarizes the inpainting results through the MSE
obtained on two different algorithms.
MSE is defined as (16):
2
1
mn
^
MSE
=
u i j
(, )
−
u i
(, )
j
(16)
mn
==
×
i
11
j
