Hi all,
I wonder if preprocessing of data like contrast stretching or pca improve the quality of classification. pca can save processing time to reduce the number of bands but is it useful when I have only three bands?
Is the separability of classes enhanced with these methods?
Best regards, Rolf
Hi Rolf,
A technique often used to remove noise in data is PCA. Even with three bands you may be able to identify significant noise and remove this through reverse PCA. Just one of many techniques, but a robust one.
James
Hi James,
Thanks for your answer. I have an aerial image with little noise I guess. In the meantime I made a PCA and used the first and second component for a ML classification. The results are better than those from the original image. Some classes are possibly better separable in the transformed space.
Concerning the noise: Is the noise reduced because you use only e.g. first and second principle component for the reverse PCA?
Rolf
Yes, in theory you can use PCA for noise reduction. If you can review the tutorial 3-3, there is a section that takes you through the process. It will be good to experiment. But if you are getting better classification results, that is what counts.
James