How can I identify and fill outliers in a 2d matrix?

Question

Sean Morgan 2024-7-24，14:38

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2139896-how-can-i-identify-and-fill-outliers-in-a-2d-matrix

评论： Sean Morgan 2024-7-25，17:45

Hello,

I am wondering if there is a function to identify and fill outliers in a 2d matrix. I know that the filloutliers function does work on a matrix, but it only works on each column separately. I am hoping to look at outliers in both directions.

There is a function fillmissing2 that does exactly this, but for missing values rather than outliers. Does anyone know if there is such a function? Or perhaps a workaround?

Thanks in advance,

Sean

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2024-7-24，14:55

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2139896-how-can-i-identify-and-fill-outliers-in-a-2d-matrix#answer_1490041

编辑：Matt J 2024-7-24，14:57

在 MATLAB Online 中打开

You could detect and replace the outliers with missing values by doing,

outlierMap = abs(yourMatrix-medfilt2(yourMatrix,[M,N]))>threshold; %detect outliers
yourMatrix(outlierMap)=nan; %Mark outlers as missing

and then use fillmissing2.

In addition to fillmissing2, there are also a number of offerings on the FIle Exchange for missing data inpainting, e.g.,

https://www.mathworks.com/matlabcentral/fileexchange/4551-inpaint_nans

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Sean Morgan 2024-7-25，17:45

Thank you, that is very helpful. A quick and easy work around

请先登录，再进行评论。

Answer 2

Image Analyst 2024-7-25，14:55

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2139896-how-can-i-identify-and-fill-outliers-in-a-2d-matrix#answer_1490641

在 MATLAB Online 中打开

You forgot to attach your data so it's hard to help you. Make it easy to help you, not hard.

One possibility is to use a modified median filter like @Matt J suggested. There are other methods if you can treat your data as an image. Some are easy and some are more complicated depending on what the outliers look like with respect to the surrounding "good" values.

If you have any more questions, then attach your data and code to read it in with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

As an example here is a demo to remove salt and pepper noise from an image.

% Demo to add "salt and pepper" noise to a gray scale image, then restore the image by removing this noise

% with a modified median filter that acts only on the noise pixels

% and not upon "good" non-noise pixels.

clc; % Clear command window.

clear; % Delete all variables.

close all; % Close all figure windows except those created by imtool.

imtool close all; % Close all figure windows created by imtool.

workspace; % Make sure the workspace panel is showing.

fontSize = 15;

% Read in a standard MATLAB demo image.

folder = fullfile(matlabroot, '\toolbox\images\imdemos');

baseFileName = 'coins.png';

% Get the full filename, with path prepended.

fullFileName = fullfile(folder, baseFileName);

if ~exist(fullFileName, 'file')

% Didn't find it there. Check the search path for it.

fullFileName = baseFileName; % No path this time.

if ~exist(fullFileName, 'file')

% Still didn't find it. Alert user.

errorMessage = sprintf('Error: %s does not exist.', fullFileName);

uiwait(warndlg(errorMessage));

return;

end

grayImage = imread(fullFileName);

% Get the dimensions of the image. numberOfColorBands should be = 1.

[rows, columns, numberOfColorBands] = size(grayImage);

% Display the original image.

subplot(2, 3, 1);

imshow(grayImage);

title('Original Gray Scale Image', 'FontSize', fontSize);

% Enlarge figure to full screen.

set(gcf, 'Position', get(0,'Screensize'));

% Generate a noisy image with salt and pepper noise.

noisyImage = imnoise(grayImage,'salt & pepper', 0.05);

subplot(2, 3, 2);

imshow(noisyImage);

title('Image with Salt and Pepper Noise', 'FontSize', fontSize);

% Median Filter the image:

medianFilteredImage = medfilt2(noisyImage, [3 3]);

% Display the image.

subplot(2, 3, 3);

imshow(medianFilteredImage);

title('Median Filtered Image', 'FontSize', fontSize);

%----------------------------------------------------------------------------------------------------------

% Two ways to identify the noise: (1) assume it's above or below some limits - like it's just pure 0 or pure 255

% or (2) identify it as some number of gra levels away from the median value in a neighborhood.

% Method %2 is good if the noise is not pure salt and pepper noise but has values other than 0 and 255.

% Use either method #1 OR method #2, not both. So comment one of them out!!!

%----------------------------------------------------------

% Method #1: darker or brighter than some specified levels.

darkLevel = 0;

brightLevel = 255;

% Find the noise. It will have a gray level of either 0 or 255.

noiseMap = (noisyImage <= darkLevel | noisyImage >= brightLevel);

%----------------------------------------------------------

% Method #2: darker or brighter than some specified levels from the median filtered image.

% differenceThreshold = 64; % Whatever you want.

% Find the noise. It will have a gray level more than differenceThreshold away from the median value in a neighborhood.

% noiseMap = imabsdiff(grayImage, noisyImage) >= differenceThreshold;

%----------------------------------------------------------

% Display the image.

subplot(2, 3, 4);

imshow(noiseMap);

title('Noise Map', 'FontSize', fontSize);

% Get rid of the noise by replacing with median.

noiseFreeImage = noisyImage; % Initialize

noiseFreeImage(noiseMap) = medianFilteredImage(noiseMap); % Replace.

% Display the image.

subplot(2, 3, 5:6);

imshow(noiseFreeImage);

title('Restored Image', 'FontSize', fontSize);

The top middle image is the original starting image with noise. The lower left image is just a map of the noise that was identified. The final lower right image is the image with those noise regions replaced with the median filtered image. So, this is a modified median filter algorithm. If your outlier regions are very large then you might have to use regionfill, instead of the median medfilt2, to basically "smear in" good surrounding values into the outlier/noise region (some people call this "inpainting").

Again, attach your data if you want more/better help.