Image Processing Reference
In-Depth Information
Demonstration of resolution conversion robustness. The upper image is a kanji
character originally scanned at 300 dpi. It has been filtered by a 5
×
5 window trained on west-
ern alphanumeric characters. The accurate results suggest that the statistics of different al-
phabets are very similar. Reproduced from Loce
2
where further details may be found.
Figure 3.12
of filtering binary images for other applications have been presented. In all cases,
the filters were derived from a training set and applied to data from that set. In prac-
tice, the process of filter training is more complex and this is discussed in detail in
the next chapter.
References
1
E. R. Dougherty and J. Barrera, “Logical Image Operators,” in
Nonlinear Fil-
ters for Image Processing
, E. R. Dougherty and J. Astola (eds.), SPIE Press,
Bellingham, WA, 1-60 (1999).
2
R. P. Loce and E. R. Dougherty,
Enhancement and Restoration of Digital Doc-
uments: Statistical Design of Nonlinear Algorithms
, SPIE Press, Bellingham,
WA (1997).
