How to apply thresholding at different parts of an image?

I have a grayscale image and location of some pixels from where i have to check whether the pixels connected to it have intensity greater than the given threshold .

There is a seperate threshold for every given starting pixel.

For example,

for a given pixel location p[i j] wih a threshold value 't' , I have to start at the pixel location [i j] in the grayscale image and check if the pixels connected to it (in 8 directions) have intensity greater than threshold 't'.If yes,then the process needs to be continued with that pixels too until no more pixels with intensity greater than threshold are found.

The process should start simultaneously from all given pixel locations in the image.

I need a logical array of 0s and 1s as result,the ones whose intensity are greater than t will be marked as 1 and the process will be continued till connected pixels don't have intensity greater than 't'.

Can somebody help me how to do this?

Answer 1

[Image Analyst]

I think this is simply a labeling problem so you can use the built-in function bwlabel().

You said you have a threshold for every single pixel in the image. OK that means you have another image where the values are the thresholds. You said you then want to threshold (binarize) all the pixels according to the threshold you have for each pixel location, so you're seeing if your image values are more than your threshold image values. And then you simply want to find all pixels above the threshold that are connected to the one at row i, column j. So you can just do

binaryImage = grayImage >= thresholdImage; % Binarize the image.
% Now identify connected components
[labeledImage, numberOfRegions] = bwlabel(binaryImage);
subplot(1, 2, 1);
imshow(labeledImage, []);
% Now determine the label at row i, column j, where i and j are known, specifed coordinate values.
ijLabel = labeledImage(i, j);
% Extract only the region with that label and display it.
ijBlob = ismember(labeledImage, ijLabel);
subplot(1, 2, 2);
imshow(ijBlob);

Answer 2

[Rik]

What you describe is a region growing algorithm, of which you can find an implementation here.

[Krishna Chaitanya]

I have tried this code using recursion but it is going into an infinite recursion.Can somebody help me where did i err.. and suggest possible changes.

Here,Ig is the grayscale image with intensities to check

i is column vector containing x-coordinates of seed pixels

j is column vector containing y-coordinates of seed pixels

threshold is column vector containing threshold values for seed pixels

reg_output is the final logical array output after region growing ,

visited is a matrix used to prevent redundant computations using recursion,

global reg_output;
reg_output = zeros(size(Ig));
global visited;
visited = zeros(size(Ig));
global row;
global col;
row = [-1;-1;-1;0;0;1;1;1];
col = [-1;0;1;-1;1;-1;0;1];
global si;
global sj;
[si,sj] = size(Ig);

for ll = 1:length(i)
    region_grow(i(ll),j(ll),threshold(ll));
end
figure,imshow(reg_output)

function region_grow(m,n,t)
global reg_output;
global visited;
global Ig;
global row;
global col;
global si;
global sj;
reg_output(m,n) = 1;
visited(m,n) = 1;
for kc = 1:length(row)
    i = m+row(kc);
    j = n+col(kc);
    if i>=1 && i<=si && j>=1 && j<=sj
        if visited(i,j)==0 && Ig(i,j)>= t
            region_grow(i,j,t);
        end
    end
end
end

[Rik]

Well, if you completely ignore my advice and want to write something yourself, go ahead. But I am not going to correct your code after having gone to the trouble of writing the implementation I posted on the FEX.

That being said, there is one important difference between your setup and a standard region growing: "The process should start simultaneously from all given pixel locations in the image." This means you need to adjust the input checking for the seed, as well as the initialization of the output. These are the only things you need to edit.

%Specifically, you need to edit the code below to support multiple seeds at once

        case 'seed'
            ME.message='The seed must be a vector with a valid position.';
            %The image itself was already checked if entered, load default
            %image if none was provided.
            IM=default.I;if isfield(options,'I'),IM=options.I;end
            try
                if numel(item)~=ndims(IM)
                    return
                end
                seed=item(:)';
                seed_=num2cell(seed);val=IM(seed_{:});%#ok<NASGU>
                %if this works, than the seed must be a valid vector
            catch
                return
            end
            ME.message='';
            options.seed=seed;

%and this line of code:
seed=num2cell(seed);result(seed{:})=true;%pre-allocate the result

Note that this can by definition only work for a single threshold per image. The only other thing you need to do is specify you want the maximum nHood.

reg_output=RegGrow(Ig,[row col],t,'nHood',8);

You don't need to change any other line of code and you can just call that last line I wrote.

[Krishna Chaitanya]

Can you please do those changes for me? I'm new to matlab so i couldn't understand the code well cause I don't know many of the functionalities and it'll take a lot of time for me.

Thank you

[Rik]

The first block is simply input checking, so you could replace it with options.seed=item; if you don't care about input validation.

The second part should be easy to do on your own. The result variable should be marked with true for each seed location.

As a general remark: everyone starts being new at Matlab and you will only learn it by putting in the time. Start with simple loops and optimize when that appears to be too slow.

[Rik]

Did you already attempt to implement my suggestions?

[Krishna Chaitanya]

I haven't yet.I will try now and keep you posted.

Thank you.