Image segmentation and counting.

Dear all, I am a BMEG student. I got a homework to count the cell of leaves by Matlab. The original figure is shown below.

And I used the following script to generate some result (I copied some code from examples).

rgb = imread('original.png'); %load image
rgb2 = rgb2gray(rgb); %change it to gray
rgb2 = imcomplement(rgb2); %invert the color
rgb2 = adapthisteq(rgb2, 'ClipLimit', 0.1); %enhance contract
BWdfill = imfill(im2bw(rgb2), 'holes'); %fill holes 
BWnobord = imclearborder(BWdfill, 4); %remove the boundary cells
rgb_perim = bwperim(im2bw(BWnobord)); %find the boundary
overlay1 = imoverlay(rgb2gray(rgb),rgb_perim, [.3 1 .3]); %overlap figure with boundary
imshow(overlay1)

bw = BWnobord;
cc = bwconncomp(bw, 8);
graindata = regionprops(cc,'basic');
grain_areas = [graindata.Area];
length(grain_areas);
min(grain_areas);
max(grain_areas);

s = regionprops(BWnobord, {'Centroid'});
hold on
numObj = numel(s) %show the total number of cell
label_num = 0;
for k = 1 : numObj
    if (grain_areas(k)> 200) & ( grain_areas(k) < 4000) %set area range
    plot(s(k).Centroid(1), s(k).Centroid(2), 'ro'); %label selected ones
    label_num = label_num + 1; %count number
    end
end
hold off
label_num

And I got the following figure.

But the two circled regions didn't have a good segmentation. Does anyone have solution for this condition?

Thanks,

Xiaoquan.

Answer 1

[Ashutosh Kumar Upadhyay]

the size of those blobs may not fit into threshold range. try the code at http://in.mathworks.com/matlabcentral/fileexchange/48610-wavelet-based-noise-robust-image-segmentation with parameters as IsObjectLighter=1; NoOfScales_J=6; KernelBandwidth_h =1; NeighbourhoodDistance_RN = 0; ContrastThreshold_Th = 1; MorphSmoothRad=0; IsGelImage=1; smallRadiusofSpot=1; and apply area filtering operation on "seg" variable.

Answer 2

[Sarah Salvino]

Hi, I was working in a similar project, could you please help me? My email adress is: [e-mail]