Extract Noise Parameters from images

Hi All:

I'm trying to extract noise parameters from a gray image which contains an object. I did the binarisation then masked the image. I did this because I have the noise just around the object, so I want to ask how can I get the parameters of the noise from the masked image 'around the object'???

Thank you in advance

Answer 1

[Image Analyst]

Maybe it has no noise. What do you consider to be noise? Do you have a noiseless "ground truth" image?

[Soum]

No The image I'm working on is a noisy image EM Microscopy and I want to know the nature of the noise in it and extract their parameters

[Image Analyst]

Can you put in a sample that is known to be uniform with no texture? Otherwise you just have an array of a bunch of numbers and there's no telling what those numbers should really be if they were free of noise.

[Soum]

I couldn't understand what you said! but overall I can't get the images because they offered to me anyway my images are noisy for sure and I want to test the noise nature

[Image Analyst]

Just quit trying to find and measure the noise and just get rid of it, if you can. For example, try the median filter, medfilt2(). That doesn't care what the noise it - it just does its thing. And you may find that it reduces the noise enough. There are other denoising methods you can use that make various assumptions and have various pros and cons, but the median filter is the easiest to try. If it works, great. If it doesn't, then you will have to try other methods. Post your image for further advice.

[Soum]

Thank you Image analyst but I don't want to denoise the images I want to know which noise is in these images in order to do a analytic studies.I found for example that I should plot the PDF of the image then analyse the distribution of the pixels right??

[Image Analyst]

No, that's not right. Why do you think that would even be close to being correct? The spread/variance/StdDev of the pixel values can only be considered the noise if the image is uniform , which you have clearly said it is not. But valid structure in the image will cause a non-zero variance even though it's not noise so you can't just say that any spread is due all to noise.

[Soum]

thank you again IA, that's why I did the binarisation then mask the image for analyse only the background where I have only the noise I'm asking how can I test the nature of the noise

[Image Analyst]

If the area that you masked and define as background is known to be uniform , then any variation there can be considered noise. If you think the noise is additive, then that noise could be considered to be everywhere in the image. Other than that, I'm not sure what you mean by "test".

[Soum]

IA the images are noisy by nature means during their acquisition they affected by noise I'm not sure if the noise in these images is a gaussian or laplacian or ...I study the distribution for example of this noise or maybe calculate the variation I don't know what I should calculate exactly as noise parameters if you can help me in this part this ll be great

[Image Analyst]

Take a histogram of the uniform parts and look at the shape of the histogram. There is a histfit() function in some toolbox.

You might also take a look at RANDRAW: http://www.mathworks.com/matlabcentral/fileexchange/7309-randraw

[Soum]

Thank you I'll see

[Asrith Pyla]

What do you mean by images being uniform?

------- Image Analyst

[Image Analyst]

I'm not an electron microscope operator so I don't know. Maybe you can use a very long exposure to get the "no noise" case while a very short exposure will result in a noisy image.