Information Technology Reference
In-Depth Information
Fig. 1. Method block diagram summarizing the whole process
intensity level higher than retina and blood vessels. Next we invert the process
and calculate the inverse pyramid (containing the RBP) up to the level-0 (origi-
nal image size). At level-0, the centroid of the RBP is calculated and, finally, this
centroid is used as central point to built the image subwindow (261x261 pixels).
A similar idea is used in [11], where the pyramid is created using a simple Haar-
based discrete wavelet transform. Equally, in [12], a first approximation to the
pixels detection belonging to the papillary area is done by applying a threshold
in each of the three RGB channels. Experimentally, the search for points be-
longing to the papillary area by this operator on the DRIONS database [13] has
produced a 100% success. The Gaussian pyramid only was applied to the image
red channel and the number of levels used was N=3. The figure 3 shows the
Gaussian pyramid of an input eye fundus image, the inverse Gaussian pyramid
(containing RBP) and the final subwindow.
2.2 Obtaining Interest Points by Laplacian Pyramid
The Laplacian pyramid was initially introduced by Burt&Adelson [14] in the
context of compression of images. However, this technology has proved be use-
ful in different image analysis tasks: generation and reconstruction, progressive
transmission, multi-scale feature detection and enhancement. Just as we may
view the Gaussian pyramid as a set of lowpass filtered copies of the original
image, we may view the Laplacian pyramid as a set of bandpass filtered copies
of the image. The scale of the Laplacian operator doubles from level to level of
the pyramid, while the center frequency of the passband is reduced by an octave.
Search WWH ::
Custom Search