I used MATLAB software to classify point cloud data into 3 classes; (1) main; (2) protrusion and (3) recess. In MATLAB I assigned a distinctive color to each class to differentiate between classes (yellow, red and blue; but black is not among classification color and it is only background color which is the places which no point exist and therefore, no class was assigned). When the classified point cloud displayed, I took a screen shot of it (TIFF images attached). I wish if I could attach point clouds (txt format), but apparently it is not possible.

In my MATLAB code, there are several parameters to adjust to get the optimum classification results (presented in three colors in reference image and reference point cloud). If I change those parameters from optimum values to other values, then the classification results are changed.

Now, if I assume the reference figure (or reference point cloud) is the base for comparison, I need to calculate how much accurate the other three figures have been classified?

For example, if I say %30 of whole reference image is in red (openings), then how much is this red percentage in any of the other three figures?

Similarly for blue and yellow color percentage. This percentage or any other quantitative comparison. I have been trying to find out a way to compare the accuracy of classification either using MATLAB or ARC GIS/MAPS, ENVI or any other tools, but so far no luck.


If I undestood correctly, what you are looking for is called an "accuracy assessment". This can be done in most GIS Software, most of the time it's to calculate the accuracy of raster classifications.

The output of an accuracy assessment is a "confusion matrix", which basically tells you, how accurate the classification of each class was. So if you got 2 classes red and yellow, it tells you how many "red" where correctly classified as "red", how many "yellow" where falsely classified as "red", and how many "red" where falsely classified as "yellow" (and so on).

Here is the implementation in ArcGIS and more information about the confusion matrix, it basically works the same in every GIS that can handle raster data.

|improve this answer|||||
  • Thank you for your comment.Yes, it is called confusion matrix and I have read about it. But the question is how to import the point cloud or TIFF images into any GIS family software? In none of manuals or the links or youtube tutorials I watched, it is mentioned how? As only maps exist as dataset for GIS family software, based on what I saw in examples. Could you please advise how can I attach the two point clouds or the TIFF images to my question, so that the ones who are expert in calculating error matrix, try the dataste/images and see if it is possible to compare? – fifi Feb 21 '19 at 21:28
  • Assuming the TIFF are GeoTIFF (TIFF data with geographical information), the data ingestion can be as simple as File -> add data (in ArcMap). As far as i know, Matlab can export data as GeoTIFF. If you have no possibility to export the Image as GeoTIFF, there is always the option of georeferencing the Image directly in ArcMap after adding the data. – Rob Feb 22 '19 at 10:23
  • Thank you for your comment Rob, but TIFF images are not georeferenced. I explained above how they were generated, I took a screen shot of displayed 2D point cloud to create a TIFF image. But classified point clouds are consist of XYZ and intensity information. Please advise how I can attach them to the post here. – fifi Feb 23 '19 at 15:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.