The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
825 
The structuring element that we use is a cross structuring 
element (4-connected). The area parameter is determined by 
computing the minimum area of connected boundaries inside 
the colon wall. The connected boundaries knowing as closed 
edges are simply derived using Marr-Hildreth operator (Marr et 
al., 1980). Marr-Hildreth operator locates edges at zero 
crossings of the image that is first smoothed with a Gaussian 
mask and then the second derivative is calculated; or we can 
convolve the image with the Laplacian of the Gaussian, also 
known as the LoG operator: 
V 2 (G®/)=V 2 G®/ (5) 
The Marr-Hildreth operator is used since it is symmetric and 
finds edges in all directions and also zero crossings of the 
second derivatives always form closed contours which we need. 
They are so simple to be determined as well; all to be done is to 
look for a sign change. 
Next, area top-hating is performed to subtract the result from 
the original image. Let the result of area top-hating be J: 
■/ = /-(/«(«)*) <6) 
After that, it performs opening operator using a structuring 
element with the same size and shape as the primary erosion 
structuring element in order to smooth contours of the image 
and eliminate false touching. 
JoH = (J@H)®H 
(7) 
where © = dilation operator 
0 = erosion operator / 
Morphological opening is then followed by a global threshold 
using Otsus' method (Otsu, 1979) in which the threshold is 
chosen to minimize intraclass variance of black and white 
pixels. This threshold is used to discard extra parts and make a 
binary version of the images passed through opening operator. 
After extraction of polyp candidates this way, their boundaries 
are simply identified by determining black pixels adjacent to 
white ones. Figure 2 illustrates an example result of performing 
AMPD and Figure 3 gives an overview of the algorithm. 
Figure2. Polyp detection a) colonic polyp specification 
on CT scan b) extracted colonic polyp by AMPD 
Figure 3. Overview of AMPD algorithm 
2.4 Classification 
Classifying candidate features as polyps and non-polyps 
completes the detection process. For polyp/fold classification 
we present a novel Template Matching Algorithm (TMA) 
which is based on Euclidean distance searching regarding that 
typical model for polyps can be assumed either spherical or 
ellipsoidal. The algorithm requires two polyp templates 
including a local window and a template pattern. One pattern is 
determined to be a circle and the other one, an ellipse (Figure 
4). 
Figure 4. Polyp templates a) circle pattern b) ellipse pattern 
The window size is considered to be equal to the largest polyp 
candidate and pattern templates' sizes are selected to be as small 
as the smallest segmented component. In order to find required 
sizes we calculate area within each segmented boundary in the 
image. Area of on pixels in an image is computed by summing 
the areas of each pixel in the image. The area of an individual 
pixel is determined by looking at its 2-by-2 neighborhood. 
There are six different patterns, each representing a different 
area: Patterns with zero on pixels (area = 0), Patterns with one 
on pixel (area = 1/4), Patterns with two adjacent on pixels (area 
= 1/2), Patterns with two diagonal on pixels (area = 3/4),