To segment nuclei without consuming a lot of computing power we are proposing simple steps as follows: Segmentation by color, Morphological operations, Modified watershed.
We started the segmentation of the blood cell image by the method of shape detection, which is based on the followig steps :after reading the image we detect the contour/outline and so the entire cell (Figure 1: b), and then we dilate the image by the structuring element disk and fill the interior gaps (Figure 1: c), after that we remove the connected objects on border and smoothed the object by the structuring element disk which represents the segmented image (Figure 1: e), and we finish by visualising the segmentation of the image (Figure 1: f).
The segmentation based on the shape of the Nucleus cell is not reliable, because, for each image, it is necessary to modify the number of times of erosion of dilation.
(a) Originale image; (b) Binary Gradient Mask; (c) Dilated Gradient Mask; (d) Binary Image with Filled Holes and Cleared Border; (e) Segmented Image; (f) Mask Over Original Image;
As we realized previously that the segmentation by shape is not an efficient method for the segmentation and reason that led to that , so we suggest a different method which is the segmentation by color , and to use this method to accomplish better results that we pursued the following steps : as always we start by reading the image of the blood cell , and then extract the blue plane and apply the binarization , after that we remove the small objects from the binary image, and finish by applying the morphological operations 'closing and opening' and visualizing the results of the segmentation.
We find the color-based segmentation of the Nucleus cell to be reliable, and apply to all images without any modification.
In order to make the nuclei more evident, we used a simple equation to extract the blue plane from RGB color space (Figure 2: b), describes as: BPlane = I_B - 0.5 * I_R - 0.5 * I_G
The histogram obtained of the blue plane extracted (Figure 2: c) is composed of two areas the first is greater than 29 represent the nucleus, the other area is smaller than 29 which represents the rest of pixel values, we chose as threshold 29 to binarize the grayscale image.
After we binarized the grayscale image, we can now remove all connected components that have fewer than 1000 pixels from the binary image (Figure 2: e).
We applied the closing operation which is the erosion of the dilation of
a set by the structuring element to remove small holes in the
foreground, changing small islands of background into the foreground.
Together with the closing, we used the opening operation to remove the
noise, by removing the small objects from the foreground of the image,
and placing them in the background (Figure 2: f).
(a) Originale image; (b) Extracting the blue plane; (c) Histogram of extracted blue plane; (d) Binarization; (e) Remove small objects from binary image; (f) Morphologically open and close image;
Now the nuclei segmentation is done, the second challenge is to separate
the touching nucleus, in this section, we are proposing a modified
watershed algorithm.
The "raw" watershed transform is known for its tendency to
over-segment an image (Figure 3: d). Each local minimum, no matter how
small, becomes a catchment basin. A common trick, then, in
watershed-based segmentation methods is to filter out tiny local minima
and then modify the distance transform so that no minima occur at the
filtered-out locations. This is called minima imposition (Figure 3:
e).
The final step is to modify the distance transform so it only has minima
at the desired locations, and then repeat the watershed steps (Figure 3:
f).
(a) Mask Over Original Image; (b) Distance Transform of Binary Image; (c) Complement of Distance Transform; (d) Watershed Transform; (e) Produce small spots that are roughly in the middle of the cells; (f) Final result
We present a model of WBC segmentation, which first detects and then segments to achieve accurate WBC segmentation. In the detection stage, we used the method of shape detection to detect the contour and the entire cell and visualize the segmentation of the image which was not an efficient method, so we suggest the segmentation by color to get better results after extracting the blue plane to apply binarization, removing small objects from the binary image and applying morphological operations which are closing operation and opening operation. After the segmentation was done, the last phase was the splitting of touching nuclei and cells using the modified watershed algorithm.