How can I estimate the variance of noise of an given image?

An input image is given, I am supposed to estimate the variance of noise of that image.

Answer 1

[Image Analyst]

Try running various filters on your image until you think you've achieved a noise free image. Then subtract it from the noisy image and pass the result in to var(). Make sure you cast your images to double because a subtraction of uint8 pixel values will clip at 0 and not give you the negative values that you need.

[shimul]

Thanks Image Analyst, actually I want to calculate the threshold value to perform soft threshold denoising on image using the calculated threshold. If I denoise the image using filters to calculate the threshold and then again try to denoise the image using soft thresholding, will it be meaning full, Sir?

Please provide me some suggestions

[shimul]

if I calculate the variance of high frequency components of the image, it will be close to the variance of noise.

Is it a proper way to find the variance of the noise.

[Image Analyst]

You can compute a low pass filtered version of the image

lowPass = conv2(double(grayImage), ones(25), 'same');

Then subtract it to get the noise only (assuming this is noise):

noiseOnlyImage = double(grayImage) - lowPass;
noiseVar = var(noiseOnlyImage);

See how that goes.

[Soum]

Thank you Image Analyst just for conv we have to use conv2 because we are working on images about the variance I obtained a vector I want to have just one value as variance, how?

[Image Analyst]

It should have been

noiseVar = var(noiseOnlyImage(:));

[DM]

hello, image analyst. Do you know how to find the variance of multiple images?

[Image Analyst]

See the code in the FAQ: The FAQ on processing a sequence of files