how to determine threshold value for separating target pixel in an image?

I need to perform target detection in an image for that I am having a target detection algorithm which performs some operation in each pixel and finally produces detector output d(t) for each pixel.Now I have to make following conditions if d(t) > threshold value,then the pixel is target otherwise the pixel is not target. My question is how to determine threshold value? Is there any method for determining threshold value?

Answer 1

[Image Analyst]

Try using my interactive thresholding app: http://www.mathworks.com/matlabcentral/fileexchange/29372-thresholding-an-image

You can take a histogram and try to think up some algorithm that works well. I think the triangle thresholding algorithm works well. graythresh(), which uses Otsu, works well if you have a clearly bimodal histogram with two large humps. Not so well otherwise.

Post a screenshot of your histogram. Attach your image.

[mari pavithra]

sorry for late reply,I completed my entire program just now only.Here's the histogram of my output.

when I used graythresh function,I got 0.0588 as my threshold but my accuracy is poor.Is there any method to determine threshold value as 0 or 0.02(accurately)?

[Image Analyst]

You need to tell your histogram routine to use more bins. Then I suggest you use the triangle thresholding method instead of Otsu, which is not good for this kind of image. A triangle method is here:

http://www.mathworks.com/matlabcentral/fileexchange/28047-gray-image-thresholding-using-the-triangle-method

The triangle method will find the "corner" like around 0.01 or 0.02, whereas Otsu won't.