13.1.14. The keypoints editor

Keypoint matching with find_keypoints() works from a saved descriptor file – the .orb file a script writes with image.save_descriptor() after extracting keypoints from a reference object (the keypoints examples in the Examples menu show the capture workflow). Real descriptors usually pick up junk: corners of the background, specular highlights, features that belong to the scene rather than the object. The keypoints editor, under Tools → Machine Vision → Keypoints Editor, is the cleanup tool.

Edit File opens a descriptor (.orb or .lbp) alongside the image it was extracted from – the editor finds the matching image file saved next to the descriptor – and draws every keypoint on the image. Select the outliers with the mouse and remove them with the delete key; checkboxes toggle the visibility of each detection octave so you can work through the scales one at a time. Save, and the trimmed descriptor is written back – the previous descriptor and image are copied to .bak files first (an earlier .bak is replaced).

Merge Files combines several descriptor files into one. Capture descriptors of the same object from a few angles and distances, merge them, and the result matches the object across more poses than any single capture would.