Image Denoising of SEM Image

Hello,

I am trying to analyse microstructural features in one of my image. Yet I have couple of problems on the image. First of all, one of my friends has denoised the image using python and he did not share the details of denosising. When I try to denoise image by myself, I cannot get close to his denosising so I am wodering what I should do and what he used for denoising so that he got such a clear image. I have added the image before and after denosising (hope you can see the zip file).

Also another question is that I was using the thresholded image to measure microstructural features like length, width and volume fraction of yellow and chunky purple ones via imageJ and python. I am wondering if I can do the same with Matlab in a better way. I know my thresholding could be better but with limited coding skills that is what I can do currently :) So next question is how I can do a better thresholding to distinguish each phases which are gray, white and black in the original image

Thank you in advance.

[Image Analyst]

Please attach the original gray scale image and the code you used to produce that pseudocolored segmented image so people can try things with it.

Answer 1

[Umar]

Hi Huseyin,

For image denoising, Matlab offers various functions like imdenoise and wiener2 that can help enhance image quality. To improve your denoising results, experiment with different denoising algorithms and parameters to match the clarity of your friend's denoised image.

Regarding microstructural analysis, Matlab provides powerful tools for image processing. You can enhance your thresholding by using techniques like Otsu's method (graythresh) or adaptive thresholding (imbinarize). These methods can help distinguish between different phases in your image more accurately than basic thresholding techniques.

By exploring Matlab's image processing capabilities and experimenting with different algorithms, you can achieve better denoising results and enhance your microstructural analysis with improved thresholding techniques.

[Image Analyst]

Did some AI engine give you this? There is no imdenoise function. There are others like medfilt2 and imbilatfilt. Also you can't enhance the thresholding using techniques like Otsu's method. That's one of the most basic elementary (dumbest) threshold methods, not some superior method. What he should do, if he wants to replicate what his friend did, is to ask his friend what he did.

[Umar]

Sorry I meant to say Denoiseimage. For your information, please refer to https://www.mathworks.com/help/images/ref/denoiseimage.html https://www.mathworks.com/help/images/ref/wiener2.html?searchHighlight=wiener2&s_tid=srchtitle_support_results_1_wiener2.

[Umar]

Here is another link to my research work, https://www.mathworks.com/help/images/ref/imbinarize.html?searchHighlight=imbinarize&s_tid=srchtitle_support_results_1_imbinarize

Answer 2

[Image Analyst]

I'm going to assume you have a binary image that you just pseudocolored to be purple and yellow. (Sorry I don't have time to unzip your files and look at them.) So what you need to use is that binary image, not the colored purple and yellow image. Using that you can use regionprops and bwferet to make various measurements.

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.

[Umar]

I have done research on your comments and I do agree because Regionprops is a MATLAB function that calculates various properties of connected components in binary images, such as area, centroid, bounding box, and eccentricity. On the other hand, bwferet is a function that computes the Feret diameter of objects in a binary image, which is the maximum caliper diameter measured within the object. By combining regionprops and bwferet, you can extract a wide range of measurements from binary images, including but not limited to size, shape, orientation, and aspect ratio of objects. These measurements can be crucial for tasks like object recognition, classification, and analysis in image processing applications. For example, you can use regionprops to calculate the area and centroid of objects in an image and then use bwferet to measure their maximum Feret diameter. This information can help in distinguishing between different objects based on their size and shape characteristics. Our goal should be working as a team and provide professional feedback and comments.