I have a multispectral satellite image (RGB and NIR band) which will serve as the reference image. I also have several panchromatic (PAN) satellite images which I need to register to the multispectral (MX) images. The geographical extent covered by MX image is very large and all the PAN images fit somewhere into the bigger MX image.

The MX and PAN images are in different co-ordinate systems, however I am converting the PAN images to the system of the MX images by gdal library.

I have worked previously on registering one PAN image to another PAN image by feature extraction and feature matching using SURF features, and I tried using this with the new problem statement as well (MX with PAN) but the results were poor.

MX images have poorer resolution hence I guess SURF algorithm is not able to detect some good features in those images.

Am I approaching the problem the right way or are the ground principles different when working with satellite images with different modalities?

| improve this question | | | | |
  • 1
    Your approach seems reasonable. The resolution difference might be problematic for feature matching though, as you said. You could try downsample the PAN images to match the resolution of the MS images, and then match those. If you are using orthorectified satellite images, you only need a rotation and translation, which might help further simplify the problem. Also, if you have only a small number of images to match, or you want to check your results, you could try out manual georeferencing, e.g. in QGIS (qgistutorials.com/en/docs/3/georeferencing_basics.html). – danscr Jul 31 '19 at 8:35
  • @danscr The point about downsampling the PAN seems good. Will try. The problem with the MX images remains there though; on running the MATLAB's SURF feature extraction individually on an MX image, not many features points are returned. As an example, on a MX image of 5000 px x 4000 px, only 11 feature points were returned. – Nancy Aug 8 '19 at 5:59
  • I haven't used SURF feature extraction on satellite images before, so I'm forced to speculate. 11 feature points seem quite few, but this might be due to the low resolution of a satellite image compared to a "regular camera" image, for which SURF is mostly used. However, if the feature points are good quality they may still be useful, just make sure to validate the result. – danscr Aug 12 '19 at 7:35

Your Answer

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

Browse other questions tagged or ask your own question.