Remove noises from image

I should remove noises from attached image. I saw gaussian noise and I tried to remove it. However there are other noises at low frequency (in my opinion) that I should remove to make the image more visible. Can I apply hight pass filter? Are there other kind of noises in your opinion? this is the code that I wrote:

im=imread('lenaNoise.jpg');
%remove gaussian noise
h=fspecial('gaussian',13,.9);
s=imfilter(im,h,'conv');
figure(5)
subplot(1,2,1),imshow(im,[])
subplot(1,2,2),imshow(s,[])

then I tried to apply hight pass filter but I got wrong

%hight pass filter
sd=im2double(s);
[f1,f2] = freqspace(31,'meshgrid');
Hd = ones(31);
r = sqrt(f1.^2 + f2.^2);
Hd(r<0.6) = 0;
im_filtered=imfilter(sd,Hd,'conv');
%im_filtered=uint8(im_filtered);
figure(6)
subplot(1,3,1),imshow(im,[])
subplot(1,3,2),imshow(s,[])
subplot(1,3,3),imshow(im_filtered,[])

Can someone help me to remove all noises from this image or suggest me how to do?

Answer 1

[Image Analyst]

Gaussian noise does not mean that you should blur with a Gaussian filter. I think you're getting the concepts mixed up. The noise is on the intensity while the filter blurs spatially. Anyway, a Gaussian, or any blur, will reduce the noise, though I think the wiener2 filter is probably a better choice as far as not blurring the original image that you want to recover.

You wouldn't want to do a high pass filter, at least not one with a fairly high cutoff frequency. Your other noise looks like a very very broad sinusoid vertically so you'd have to have a very very low cutoff frequency. You might see what it looks like in the Fourier domain. A cosine would have a couple of bright spikes in the spectrum that you could zero out. If you don't see them, then you could just try to fit the image to a sine wave and divide it out. Normally a high pass filter makes the image look noisier so if you use one by doing a convolution, to have a really low cutoff frequency, you'll need a huge window - that's why I suggested doing it in the Fourier domain. See my attached demo (which works on a different image).

[Silvia Caruso]

Ok,thank you for your answer.I tried to apply your code to my image, but I have some problem. Why you choose a threshold=10.9? Can I use the same? For the central DC spikes I selected from 240 to 274,is it correct? I wrote:

   J = wiener2(im,[3 3]);% wiener filter
    frequency_J = fftshift(abs(fft2(J)));
    % Take log magnitude so we can see it better in the display.
    amplitude_J = log(frequency_J);
    minValue = min(min(amplitude_J))
    maxValue = max(max(amplitude_J))

% Find the location of the big spikes.
  amplitudeThreshold = 10.9;
  brightSpikes = amplitude_J > amplitudeThreshold; % Binary image.
  figure(8)
  subplot(1, 3, 1),imshow(amplitude_J, []),title('wiener_filtered')
  subplot(1, 3, 2),imshow(brightSpikes);
  axis on;
  % Exclude the central DC spike.
  brightSpikes(240:274,:) = 0;
  imshow(brightSpikes);

    % Filter/mask the spectrum.
    frequency_J(brightSpikes) = 0;
    % Take log magnitude so we can see it better in the display.
    amplitudeImage2 = log(abs(frequency_J));
    minValue = min(min(amplitudeImage2))
    maxValue = max(max(amplitudeImage2))
    subplot(1, 3, 2);
    imshow(amplitudeImage2, [minValue maxValue]),title('spikes zeroed   out');
    axis on;

    filteredImage = ifft2(fftshift(frequency_J));
    amplitudeImage3 = abs(filteredImage);
    minValue = min(min(amplitudeImage3))
    maxValue = max(max(amplitudeImage3))
    subplot(1, 3, 3);
    imshow(amplitudeImage3, [minValue maxValue]),title('filtered image'));

but I didn't reach my goal. Maybe because in my image there aren't spikes like in yours. Are you sure that noise in my image is a broad sinusoid vertically? I'm not so sure, but I'm not able to find other noises. I tried to remove noise also with homomorphic filter but without any results.

[Image Analyst]

That wave is such a low frequency it would be the pixels right next to the DC spike, not away from it, so they won't show up as distinct and separate spikes. With this kind of noise you might have to use some non-linear spatial denoising method like median filter or non-local means or something. How did the image get corrupted in the first place? Or did you intentionally add noise?

[Silvia Caruso]

I don't know how this image get corrupted in the first place. Probably it is an intentionally added noise, but not from me. Using non linear spatial denoising method like median filter I make the image worse.