Image Processing Reference
FIGURE 6 (a) The original image, (b) the healthy cells masked, (c) watershed for each blob
region, (d) a small mark on each cell's centroid, (e) the cell blobs separated, and (f) the final
detected benign cells.
have been previously determined. In this case, and after the masking process, many overlap-
ping benign cells appeared as one big region according to the shape and size of the normal
cells. For that reason, the watershed process in Figure 6(c) is applied to get the optimum separ-
ation in this abnormal shape of cells. In fact, to get the optimal separation of overlapped cells,
their centroids are used to put marks on them as in Figure 6(d) . Now the cells within each
cell blob have been separated and marked, as in Figure 6(e) . Finally, by applying all the previ-
ous steps, the benign cells are contoured by gray lines, extracted to count, and distinguished
from the other distorted cells, which are also shown in the second step of part one. This de-
tection operation indicates 109 benign cells out of 180 detected cells (benign and distorted) in
the image. In fact, the remaining cells (71) may be considered to be detection errors, distorted
cells (sickle-cell Anemia), platelets, or even WBCs. The image in the figure does not have any
WBCs, but if WBCs exist in other images, the proposed algorithm can easily detect and count
FIGURE 7 The detected WBCs.
Additionally, the algorithm has identified 177 cells, representing a 99.98% success ratio ac-
of the 71 nonbenign cells are currently or potentially sickle cells. First, all 71 strange shapes
(crescent, elliptic, platelets, and unknown) are discovered and displayed using the previous
steps of benign cell detection. Figure 8 illustrates the last two steps of the proposed algorithm,
which is applied to the distorted cells.