maxFeatures seems to have a geometric bias

steevo87 / thermalvis

Cross-platform, OpenCV-based functionality for image processing and computer vision in thermal-infrared

BSD 3-Clause "New" or "Revised" License

124 stars 63 forks source link

maxFeatures seems to have a geometric bias #64

Closed steevo87 closed 10 years ago

steevo87 commented 10 years ago

Perhaps at present it is just cutting the vector at that point, when instead, it should look at the weakest feature score that can get into the top "maxFeatures", and randomly sample all features of that score to fit within the quota.

To test, set the feature detector sensitivity really low, disable all tracking (and matching) and observe the distribution of features with various maxFeatures levels.

steevo87 commented 10 years ago

Note: have moved towards standard C++ vector sort operation. Vector is now sorted in ascending order. This may affect the effectiveness of subsequent filters that assume a certain direction, e.g.

reduceEdgyFeatures()
violatesProximity routine
addProjectionsToVector()
concatenateWithExistingPoints()

However, I think this has all been accounted for now.