Clean noisy data from images

Hey there,

I've been working with some images, averaging them out and plotting against depth (check out the code below). I'm trying to tidy up the data by getting rid of points that don't fit an exponential curve I've fitted to it. I've tried a couple of methods like filloutliers and sgolayfilt, but they haven't been working out too well. I used polyfit(depth_Sony, log(img_B_Masked_avg), 1) to fit the blue ('B') channel averages but the fit is not good because of the points that are off.

Any suggestions on a better approach? Thanks a bunch!

figure;
semilogy(depth_Sony, img_gray_Masked_avg, 'ko', 'MarkerSize', ms);
hold on;
semilogy(depth_Sony, img_R_Masked_avg, 'ro', 'MarkerSize', ms);
semilogy(depth_Sony, img_G_Masked_avg, 'go', 'MarkerSize', ms);
semilogy(depth_Sony, img_B_Masked_avg, 'bo', 'MarkerSize', ms);
xlabel('depth');
ylabel('mask avg value');
xlim([0 10]);
ylim([0.5*10^-1 10^1]);

[xingxingcui]

Based on your description, image denoising the code you posted I don't see any direct relationship.

Actually MATLAB already contains many methods, traditional and deep learning cases are as follows:

• noise removel
• Remove Noise from Color Image Using Pretrained Neural Network

[alberto tonizzo]

Thank you @cui,xingxing the question is more about denoising this particular dataset. Notice that the variables "img_R_Masked_avg" are vectors, not images.

Answer 1

[Mathieu NOE]

hello

this is a simple exponential fit using polyfit

I had to do a bit of manual tweaks first to slect the appropriate data

here an example on the "blue" curve

figure;
ms = 2;

ind = depth_Sony>2.5 & depth_Sony<8.5; % select valid x range (avoid large clouds)
xx = depth_Sony(ind);
yy = img_B_Masked_avg(ind);

[yy,k] = rmoutliers(yy, 'movmedian', 300, 'ThresholdFactor', 2); % remove large dips 
xx(k) = [];

[b,m] = myexpfit(xx,yy); % see function below
img_B_Masked_fit = b*exp(depth_Sony*m);
semilogy(depth_Sony, img_B_Masked_avg, 'bo',xx, yy, '*r',depth_Sony, img_B_Masked_fit, 'g--', 'MarkerSize', ms);
TE = sprintf('C = %0.2fe^{%0.3ft}',b, m);
legend('raw data','extracted data',TE);

xlabel('depth');
ylabel('mask avg value');
% xlim([0 10]);
% ylim([0.5*10^-1 10^1]);

%%%%%%%%%%%
function [b,m] = myexpfit(x,y)
% exponential fit using polyfit
    P = polyfit(x, log(y), 1);
    m = P(1);
    b = exp(P(2));
%     yfit = b*exp(x*m);

end

[alberto tonizzo]

Thank you! That looks good!

[Mathieu NOE]

tx !

as always, my pleasure !