Dear RG-Community,
for one of my projects I would like to classify multispectral images recorded at the same location over a couple of month. Now, for specific bands, the underlying reflectance signatures differ extremely. I can exclude that the difference is a real effect or due to faulty calibration. If so, would it be valid to match histograms of the images? I am just wondering as it will change the reflectance tremendously by harmonizing them. Seems extreme to me. Also, which signature to choose as the baseline?
Thanks for your input.
René
Dear Rene Heim
For your case, histogram information can be conducted for prior information. Even, the histogram value can become vector feature if you use machine learning technique either supervised or unsupervised.
With Regard
Toto Haryanto
Thanks Aparna, I am using R and know how to perform the task whether it it might be a wavelet transformation or histogram matching. My question is if matching or harmonizing multispectral imagery is valid as the matching/harmonizing appears to cover up any spectral pattern that might be relevant.
Well Rene, to fuse or blend histogram matching makes sense , for classification then I would think going only with histogram would not be all that accurate.
Do you have number of channel -information say 3 in case of RGB or more depending on the camera acquisition , again this is for look the discrete spectral info !!!
Article Histogram Curve Matching Approaches for Object-based Image C...
Dear Rene,
Of course histogram matching can cover up any spectral pattern. If for example you have a clear and hazy image (i.e. Aerosol) in that case the clear image surface reflectance can be determined by removing Rayleigh scattering, minimal aerosol scattering, and gases absorption. In that case histogram matching can be used to correct the hazy image from corrected clear image. Both images should represent the same landscape otherwise it will not work.
Best regards
Rene,
Use Idrisi and u will be able to use all the image reflectance data rather than matching them. A whole lot of spatial variation inherent in your study area will be lost if you ignore this.
Also try a soft classification or hybrid classification approach.
Cheers.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing