how to get the threshold automatically when converting image from hsv to bw

Hi! I have to convert my image from HSV to BW to do all the analysis. But I have a lot of images and I can't adjust the threshold one by one. I have checked some of my images, and the threshold change from 0.1 to 0.4. SO the change is quite large and I can't use a single threshold for all the image.

So does somebody know how can I generate the threshold for each image without manually checking the histogram? I need to choose the threshold after the first pick in the histgram.

I asked a question related to this one before. More info: https://stackoverflow.com/questions/47024527/how-to-segment-the-colony-when-i-cant-use-bw-image

Some of my codes are here:

b=imread('test4.jpg')
hsvIm = rgb2hsv(b) % convert the image to hsv format
imshow(hsvIm)
histogram(hsvIm(:,:,1))
%thresholding on the hue space
bwIm = hsvIm(:,:,1) < 0.3 % the threshold can change from 0.1 to 0.4

I also attach some pictures here:

Thank you in advance!!!

[Walter Roberson]

Sorry, No. We do not have any idea how you are judging whether part is foreground or background based on the hue. And why are you not taking into account that hue is an angle, so hue values "just before" hue 0 would have high hue ?

[Image Analyst]

We cannot invent some adaptive automatic threshold algorithm for you, especially not having seen any of the images. It could be virtually anything -- how could we possibly know? You'll just have to look at the histograms and colorcloud and figure out something.

[Ziming Zhong]

@Walter Roberson what do you mean "hue values just before hue 0 would have high hue"? Sorry that I'm quite new on Matlab and I dont really understand what you mean. Could you please explain it a bit? Thanks!

[Ziming Zhong]

@Image Analyst Not sure whether you could see this comment by @. But anyway, I just want to say that I edit a bit of my post and add some pics. Also, I thought you have that `grayshreshold` function in MatLab, and so I thought maybe you know how to do the same for the hue threshold.

[Walter Roberson]

If you look at this picture, you will see that values that you would identify as "red" happen both for hue near 0 and for hue near 360. If you were trying to segment by "red" then you would need to include hue less than some value and hue greater than some value. Low hue is on the "red to yellow" stretch, and high hue is on the "magenta to red" stretch. Red with a touch of yellow would have a low hue; red with a touch of magenta would have a high hue.

[Akira Agata]

How about trying the Color-Based Segmentation Using K-Means Clustering ?

[Ziming Zhong]

@Walter Roberson Thanks for the clarification! Its really helps! So now you mean I should choose a range of hue value according to the color of the colonies? So that I can segment the colonies apart from the background? Or what do you mean in your first comment that hue is a angle? Thanks!!!

[Guillaume]

what do you mean in your first comment that hue is a angle?

The HSV colour space is a cylinder where hue is the angular dimension (and S the radius, and V the height along the cylinder). Matlab encode all of these in the range 0-1, which for the hue maps to 0-360 degrees.

[Ziming Zhong]

@Guilaume Thanks! Yeah, I think I understand this point. That's why I convert the image from RGB to HSV to use the unique hue for each colony. But my question asked more like how can I use hue as a variable to segment the colony? I guess I already did it by converting the image, right? Or do I understand Walter or you wrong?

[Guillaume]

Walter point is that if you want to segment according to a red hue you need to consider hue values near 0 and hue values near 1 (360 degrees) since red is at both ends.

I would add to that that segmenting just using the hue is risky, you also ought to look at the saturation and value. Compare these three images:

figure;
subplot(1, 3, 1); imshow(hsv2rgb(repmat(cat(3, 0.2, 1, 1), 100, 100)));
subplot(1, 3, 2); imshow(hsv2rgb(repmat(cat(3, 0.2, 0.01, 0.01), 100, 100)));
subplot(1, 3, 3); imshow(hsv2rgb(repmat(cat(3, 0.2, 0.01, 0.99), 100, 100)))

They all have the same hue (0.2 == yellowish), yet only one of them look yellow. The other two have very low saturation (0.01) and vastly different values (0.01 and 0.99)

I would actually follow Akira's advice to use k-means clustering in the L*a*b* colour space.

[Ziming Zhong]

@Akira Agata @Guillaume Thanks Akira and Guillaume! K-Means doesn't work well either, as it can't distinguish the yellowish colony with yellowish media, even though the colony is dark. I don't really see why. I'm also thinking to use edge detection, but doesn't work at all. I don't know whether its somehow I got something wrong.

Answer 1

[Ziming Zhong]

Hmm, I just want to update my question here.

Finally, I don't need to set the threshold separately for each image. I just converted the image from RGB to HSV and then to YCBCR. After the second conversion, the threshold for YCBCR to BW is more or less the same for every image, no matter how different the colony color and size. I'm not really sure whether it is proper to do two conversions. But at least it works.

I also tried to convert RGB directly to YCBCR, but the result is not as good as through two conversions.

The only problem that I still have is that imfindcircles() on bw image doesn't really work well even though I played with the sensitivity and EdgeThreshold, and I don't know why. Do somebody know the possible reason?

[Ziming Zhong]

I guess its better i list my codes here:

hsvIm = rgb2hsv(a); % convert RGB image a to hsv format; 1st convertion
ycb = rgb2ycbcr(hsvIm); % convert to ycbcr format; 2nd convertion
imshow(ycb);
histogram(ycb(:,:,1)); % check the histogram to decide the threshold 
bwIm = ycb(:,:,1) < 0.35; % I had tested 25 images by using this threshold, and all of them resulted in perfect masks.

I started to try more image format conversions after the suggestions and finally can find this solution. Thank everybody who gave me suggestions on this question!!!

Answer 2

[Image Analyst]

Since they seem to all be circular, have you tried imfindcircles()?

[Ziming Zhong]

Yeah. imfindcircles() is the first method I used to find the colonies, as I worried about various thresholds in image conversion and so I just want to avoid it. But imfindcircle() didn't work well for some colonies, and i really have no idea why. I asked and discribed my colony problem in imfindcircles() here: stack overflow question

I have tried to adjust the sensitivity in imfindcircles(), and no idea why it didn't work at all in my black colony image. Since the accepted answer under that question, I started to consider image conversion.

Now I'm thinking about to choose the local min in the histogram of each HSV image. But I don't know whether it will work.

[Image Analyst]

Have you tried a top hat filter or bottom hat filter?

[Ziming Zhong]

Yeah, just tried. Also don't work on the black colonies at all. Also I got a problem when using imtophat(original,se), where the pixel of se set to lower than 100 the whole image turned black, while se was >100 then the whole image is quite ugly. It doesn't help even i use imadjust() later.

se = strel('disk',30)
tophatFiltered = imtophat(grey,se)
contrastAdjusted = imadjust(tophatFiltered)
[centersDark, radiiDark] = imfindcircles(contrastAdjusted,[30 60],'ObjectPolarity','dark','Sensitivity',0.925,'Method','twostage','EdgeThreshold',0.1)
h = viscircles(centersDark,radiiDark)

[Image Analyst]

Exactly what regions are you trying to find? The whole bullseye, or just the center, or something else?

[Ziming Zhong]

Yeah, the whole bullseye.

Answer 3

[Image Analyst]

Try this:

clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 25;

%===============================================================================
% Get the name of the image the user wants to use.
baseFileName = 'lab_try.jpg';
folder = fileparts(which(baseFileName)); % Determine where demo folder is (works with all versions).
fullFileName = fullfile(folder, baseFileName);

% Check if file exists.
if ~exist(fullFileName, 'file')
  % The file doesn't exist -- didn't find it there in that folder.
  % Check the entire search path (other folders) for the file by stripping off the folder.
  fullFileNameOnSearchPath = baseFileName; % No path this time.
  if ~exist(fullFileNameOnSearchPath, 'file')
    % Still didn't find it.  Alert user.
    errorMessage = sprintf('Error: %s does not exist in the search path folders.', fullFileName);
    uiwait(warndlg(errorMessage));
    return;
  end
end

%===============================================================================
% Read in demo image.
rgbImage = imread(fullFileName);
% Get the dimensions of the image.
[rows, columns, numberOfColorChannels] = size(rgbImage);
% Display the original image.
subplot(2, 2, 1);
imshow(rgbImage, []);
axis on;
caption = sprintf('Original Color Image, %s', baseFileName);
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo();

% Set up figure properties:
% Enlarge figure to full screen.
set(gcf, 'Units', 'Normalized', 'OuterPosition', [0 0.05 1 0.95]);
% Get rid of tool bar and pulldown menus that are along top of figure.
% set(gcf, 'Toolbar', 'none', 'Menu', 'none');
% Give a name to the title bar.
set(gcf, 'Name', 'Demo by ImageAnalyst', 'NumberTitle', 'Off')

drawnow;
hp = impixelinfo(); % Set up status line to see values when you mouse over the image.

% Compute HSV image.
hsvImage = rgb2hsv(rgbImage);
saturationImage = hsvImage(:, :, 2);
valueImage = hsvImage(:, :, 3);

% Display the image.
subplot(2, 2, 2);
imshow(saturationImage, []);
axis on;
caption = sprintf('Saturation Image');
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo();
drawnow;

% Create the binary image.
binaryImage = saturationImage > 0.22 & valueImage > 0.1;
% Display the image.
subplot(2, 2, 3);
imshow(binaryImage, []);
axis on;
caption = sprintf('Binary Image');
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo();
drawnow;

% Take largest blobs only
binaryImage = bwareaopen(binaryImage, 20);
% Fill them to get rid of noise.
binaryImage = imfill(binaryImage, 'holes');
% Display the mask image.
subplot(2, 2, 4);
imshow(binaryImage, []);
axis on;
caption = sprintf('Binary Image');
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo();
drawnow;

props = regionprops(binaryImage, 'Area');
allAreas = sort([props.Area], 'descend')

[Ziming Zhong]

Thank you very much for the codes! I had tried your code through 25 images containing colonies (so far I have) in different shapes and colours. On some of the images, this code works perfectly. But for some colonies, the code still got problems and I guess still the large range of thresholds among images caused problem. Comparing the results from this code with results from two image format conversions (as I mentioned above), I think I will try the two image conversions method to analyse all my colonies.

Thank you very much again for the codes!!!!!