Histogram and automatic thresholding

Hello everyone, sorry for requesting another. Really new to imaging.

I have perform the normalized gradient magnitude. Also found the histogram of it. But I wish to apply an authomatic thresholding technique based on the histogram of the normalized gradient magnitude. All edge map pixel with values greater than the threshold retains their original values, while those edge map pixels with values less than the threshold had their values set to zero.

please find my code;

im=dicomread('Image');

    %applying sobel filtering
    P1=fspecial('sobel');

    P2=P1;
    Data1=imfilter(im,P1,'replicate');

    Data2=imfilter(im,P2,'replicate');

    % gradient magnitude

    grad=sqrt((Data1.^2)+(Data2.^2));

    figure,imshow(grad,[]);

    %%Normalize the Image:

    myImg=grad;

    myRange = getrangefromclass(myImg(1));

    newMax = myRange(2);

    newMin = myRange(1);

    myImgNorm = (myImg - min(myImg(:)))*(newMax - newMin)/(max(myImg(:)) - 
    min(myImg(:))) + newMin;

    figure,imshow(myImgNorm,[]);

    %%calculating the histogram of normalized gradient

    bin=255;

    imhist(double(myImgNorm(:)),bin);

    %figure,plot(h);

The next step I need your help is:

To perform the automatic threshold based on the histogram of the normalized gradient magnitude.

input image=Normalized gradient magnitude.

Thank you for your help in advance.

Regards,

[Matt J]

It would help if you format your code in a separate font from your text using the

formatting button. This would make the post more readable.

[Josh]

Sorry Matt , I will try to format the code as you suggested.

[Image Analyst]

You didn't do it correctly. Just format the code, not your comments. And get rid of empty lines between code. Then just highlight the code (only, no regular text from you) and click the {}Code button.

Answer 1

[Image Analyst]

I don't know what algorithm you're going to use to find the threshold from the histogram. There are many, many methods. The most famous is Otsu's but I find that is only good for high contrast bi-modal histograms. An edge image may have a skewed histogram that looks kind of log-normal in shape (like a skewed Gaussian). For those kinds of histograms, I find the triangle method works fairly well, where it will find the "corner" in the skewed slope of the histogram mode.

Anyway, whatever method you select, once you have the threshold, to binarize your image, you simply do

binaryImage = yourEdgeImage > thresholdValue; % can use < or <= or >= if you want.

[Josh]

Thanks Sir for your response. I have used Otsu's method as you mentioned, I had some difficulties. But I am not familiar with the triangle method, I wish to try it as well ,please Sir can you help. I don't know if my code was right the manner I find the threshold value from the histogram using.

imhist(double(myImgNorm(:)),bin); % histogram of the normalized gradient

thresholdValue=graythresh(imhist(double(myImgNorm(:)),bin); % threshold value from histogram

[Image Analyst]

I can't help more than that now - I'm packing to go away for the weekend in a few minutes.