Clearly Identifying circular regions on a chip in a noisy environment

Hey everyone

As the summary suggests, I have been working with chip images in hopes of clearly identifying the circles via pre-processing so that I can binarize the image and use regionprops on them afterwards. I haven't had much success and any help would be much appreciated. I have shared some photos that I am working with that should help!

My current algorithm is very slow but also not very good at identification.

[Image Analyst]

Is it the small dark spots in the rectangular array that you want to "identify"? What does "identify" mean to you? Do you just want the location of the centroids of them or do you also want other attributes, like area, perimeter, diameter, etc. There is going to be some sort of cutoff because the spots gradually fade into non-existence so you're going to have to make some kind of judgment call as to what is a spot and what is noise or just part of the background.

See my Image Segmentation Tutorial in my File Exchange:

Image Analyst's File Exchange

It's a generic, general purpose demo of how to threshold an image to find blobs, and then measure things about the blobs, and extract certain blobs based on their areas or diameters.

[Bera]

Basically just find their centroids and have them clearly identified by regionprops

[Bera]

I tried your stuff Image Analyst but the images I have are a bit more difficult. I have shared them in my post!

[Bera]

@image analyst do you offer like private tutoring classes by any chance?

Answer 1

[Balavignesh]

Hi Bera,

It is my understanding that you would like to identify circular regions on a chip. It can be challenging, especially in a noisy environment. The key to improving both the speed and accuracy of your algorithm lies in effectively preprocessing the images to enhance the features of interest and suppress the noise.

• Noise Reduction: You could start with noise reduction to make the subsequent steps more effective. Gaussian blur or median filtering can be effective, depending on the type of noise present in your images. Gaussian blur helps in smoothing the image and is effective for Gaussian noise, whereas the Median Filter is effective for salt-and-pepper noise.

imgFiltered = imgaussfilt(originalImage, sigma);
imgFiltered = medfilt2(originalImage, [kernelSize kernelSize]);

• Edge Detection: After noise reduction, use edge detection to outline the boundaries of the circles. The Canny edge detector is commonly used for this purpose due to its robustness.

edges = edge(imgFiltered, 'Canny');

• With the edges detected, apply the Circular Hough Transform to identify circles in the image. MATLAB's 'imfindcircles' function is based on the Hough Transform and is particularly suited for this task. The 'ObjectPolarity' parameter should be set according to whether the circles are brighter or darker than the surrounding pixels.

[centers, radii] = imfindcircles(edges, [minRadius maxRadius], 'ObjectPolarity','bright', 'Sensitivity',0.9);

• Once you have the circles identified and their parameters (centres and radii), you can create a binary mask of the detected circles to isolate them from the rest of the image. Then, use 'regionprops' to analyze their properties

Fine-tuning the preprocessing steps and the parameters of the algorithms can significantly impact both the speed and accuracy of your circle detection.

Kindly have a look at the following documentation links to have more information on:

• Gaussian Blur: https://www.mathworks.com/help/images/ref/imgaussfilt.html
• Median Filters: https://www.mathworks.com/help/images/ref/medfilt2.html
• Edge Detection: https://www.mathworks.com/help/images/ref/edge.html

Hope that helps!

Balavignesh

[Bera]

Hey Balavignesh I really appeciate it. I'm quite new to Matlab would you be able to provide me with a bit more help? Like how the code strcuture may look its a big ask and if it's not possible I completely understand