Sub-images were collected from the the original image near the corners or the middle of the four sides using some knowledge about the original image such as image size, resolution etc. This knowledge will also be used for determining the parameters for morphological operation. Colour images were treated by just taking the red component of the RGB band, as most of the fiducial marks appear to be red in the colour image.
The above described method were tested on 14 images. Figures 3, 4, 5 showed different example images for segmenting the circular shaped fiducials. The parameters used for this kind of images were all the same. The area attribute used here was of size 100, and the ratio parameter set in our algorithm was 1.2. The result is a grey scale image of the fiducial mark.
Figure 6 shows the result of obtaining the fiducial of
the shape of a broken `` 
''. The size of the median filtering
kernel used in the procedure is 7 7.
Figure 3:
The process of the automated segmentation of fiducial marks.
(a) one of the original sub-images containing a fiducial mark;
(b) morphology opening by area 100;
(c) morphology opening by aspect ratio 1.2;
(d) subtraction of image in (c) from the image in (b); and
(e) the brightest object by using morphological reconstruction.
Figure 4:
The process of the automated segmentation of fiducial
marks for a different image (with letters).
(a) one of the original sub-images containing a fiducial mark;
(b) morphology opening by area 100;
(c) morphology opening by aspect ratio 1.2;
(d) subtraction of image in (c) from the image in (b); and
(e) the brightest object by using morphological reconstruction.
Figure 5:
The process of the automated segmentation of fiducial
marks for a different image.
(a) one of the original sub-images containing a fiducial mark;
(b) morphology opening by area 100;
(c) morphology opening by aspect ratio 1.2;
(d) subtraction of image in (c) from the image in (b); and
(e) the brightest object by using morphological reconstruction.
Figure 6:
The process of the automated segmentation of fiducial
marks for the broken `` '' shape.
(a) one of the original sub-images containing a fiducial mark;
(b) median filtering by kernel size of 7 7;
(c) subtraction of (b) from (a);
(d) threshold of (c) after histogram equalization;
(e) attribute opening by using size/ratio/orientation ( 
);
(f) attribute opening by using size/ratio/orientation ( 
);
(g) logical ``OR'' of image (e) and (f); and
(h) morphological reconstruction of (a) using (g).
Figure 7 shows the results of each step for detecting réseau marks in a grey-scale image.
Figure 7:
The process of the automated segmentation of cross shaped fiducial marks.
(a) the original image;
(b) the difference between the original image and the median
filtered image by a kernel size 1 5;
(c) the difference between the original image and the median
filtered image by a kernel size 5 1;
(d) morphology opening of (b) using area attribute of size 20;
(e) morphology opening of (c) using area attribute of size 20;
(f) the logical 'AND' image of (d) and (e);
(g) morphology reconstruction using (d) and (f);
(h) morphology reconstruction using (e) and (f); and
(i) the logical 'OR' image of (g) and (h).