remove outliers in 3D point data

Hi guys, I want to thank you in advance for you help. I have a point cloud of an object, separated by axis:

    points = [x, y, z];

which when plotted on a scatter contain some outliers that I want to get rid of.

I've already got a distance matrix of the point set using the following code:

distanceMatrix = squareform(pdist(points, 'euclidean'));

I can use logical indexing to remove distances beyond a certain standard deviation (I'm using 1 here),

newDistanceM = distanceMatrix>mean(mean(distanceMatrix))+1*std(std(distanceMatrix));

Where I am having trouble is using newDistanceM to filter out the points that are outliers in my original set. It's probably really simple to do this but I'm not seeing it.

Any help on this is, of course, greatly appreciated and if anybody has a better/more robust way of achieving what I describe, I very much welcome the feedback.

Cheers

[Baltam]

Are you sure you are calculating the correct distance? If you need the distance of each point to the origin than you should use:

points = rand(50,3);
distance = sqrt(sum(points.^2,2));
indices = distance<(mean(distance)+std(distance)); %note I use smaller than instead of bigger
remainingPoints = points(indices,:);

[jay]

thanks for this comment, it's helped me correct what I was doing. Cheers!

Answer 1

[Image Analyst]

Brett uploaded an outlier removal tool that uses Grubbs test: http://www.mathworks.com/matlabcentral/fileexchange/3961-deleteoutliers

I tend to like the median absolute deviation. Compute distances from the centroid (mean x, y, z) and get rid of points where the distance is greater than some factor times the MAD value.

Attached is an example of how to use MAD to get rid of Salt and Pepper noise.

[jay]

Thanks for the attached file, it was very infromative!!