Image Processing Reference
In-Depth Information
Both designs were tested and the partial-sort reached a maximum frequency of
80Mhz, while the histogram was capable of working at frequencies over 100 MHz.
When the performance was measured overall (i.e., taking a video stream from
the PC through the FPGA and back to the PC), the partial-sort algorithm achieved
13.1 frames per second (fps) while the histogram approach achieved 9.2 fps. The
limiting factor was the overhead introduced by moving the data in and out of the
FPGA device.
Another aspect to take into account is that the performance of the partial-sort
design depends on the filter to be applied and the input stream, whereas the histo-
gram performance is not affected by the filter nor the input stream and has the ad-
vantage of being adaptive to window sizes. The design is so small that it can easily
be replicated on the same device to improve the performance.
9.5 Summary
This chapter has presented an overview of how the techniques introduced in the
topic may be used to solve a real-world problem. The use of a training algorithm is
key to the creation of an appropriate nonlinear filter. In this case, the training set
was produced by hand on a small image and this was used to obtain the optimal soft
morphological filter. The filter itself was designed using a genetic algorithm run
over 500 iterations. Two different quality criteria were compared. The resulting fil-
ters produced images with the noise significantly reduced and the structure intact.
The second half of the chapter considered approaches to hardware implementa-
tion of soft morphological filters in real time. Single images may be processed eas-
ily in software, so the presented example considered spatio-temporal images. It
presented two approaches and gave performance metrics for each.
References
1
G. E. Brueckner, R. A. Howard, and M. J. Koomen, “The large angle spectro-
scopic coronagraph (LASCO),”
Solar Physics
,
162
, 357 (1995).
2
S. W. Kahler, “Solar flares and coronal mass ejections,”
Annual Reviews of As-
tronomy and Astrophysics
,
30
, 113 (1992).
3
L. Koskinen and J. Astola, “Soft morphological filters: A robust morphological
filtering method,”
J. Electron. Imaging
,
3
, 60-70 (1994).
4
M. S. Hamid, S. Marshall, and N. Harvey, “GA optimisation of multidimen-
sional gray-scale soft morphological filters with applications in archive film
restoration,”
IEEE Trans. Circuits and Systems for Video technology
,
13
(5),
406-416 (2003). See also
13
(7), 726 (2003).
5
P. Kraft, N. Harvey, and S. Marshall, “Parallel genetic algorithms in the opti-
mization of morphological filters: a general design tool,”
J. Electron. Imaging,
6
(4), 504-516 (1997).
