I have used around 200 sqkm high resolution single band colour Bing image for object based image analysis on ecognition developer. But due to different tile based exposure made it worse to classify. the segmentation goes different for two areas. Can anyone suggest how to pre process the image before segmentation. I have added an example of exposure problem through image. [different tile based exposure problem]

I have used ecognition developer provided equalizing tools like Gamma Correction, Histogram, linear equalizer and standard deviation. But nothing works.


Those two images are very different and as such, it will not be simple to apply the same segmentation and classification on both areas. In general, working with imagery from a WMS system, like Bing, is not recommended, as that limits you to fewer bits of information than the imagery was acquired with, which limits your options for adjusting preprocessing and cross-calibration. (Also note that using the imagery from the WMS as input to your model may be against the ToU of the Bing imagery - see the Bing ToU.)

One option would be to simply create two different classifications in the different sections, since you will never be able to fully normalize the two areas.

Another would be to acquire the source data and see if you can normalize those datasets better yourself.

Or you could potentially look at the overlap between the two source datasets and use that to cross-train the classification and segmentation methods.

  • ok, thank you for reminding Bing ToU. I will modify my project with other available option. – Bappa May 11 '20 at 14:07

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.