match_features.py

#!/usr/bin/env python3

import numpy as np

import pyboof as pb
from pyboof.swing import visualize_matches

# Load two images
image0 = pb.load_single_band("../data/example/stitch/cave_01.jpg", np.uint8)
image1 = pb.load_single_band("../data/example/stitch/cave_02.jpg", np.uint8)

# Set up the SURF fast hessian feature detector.  Reduce the number of features it will detect by putting a limit
# on how close two features can be and the maximum number at each scale
config_fh = pb.ConfigFastHessian()
config_fh.extractRadius = 4
config_fh.maxFeaturesPerScale = 300

# Create the detector and use default for everything else
feature_detector = pb.FactoryDetectDescribe(np.uint8).createSurf(config_detect=config_fh)

# Detect features in the first image
locs0, desc0 = feature_detector.detect(image0)
locs1, desc1 = feature_detector.detect(image1)

print("Detected {:4d} features in image 0".format(len(desc0)))
print("         {:4d}             image 1".format(len(desc1)))

config_greedy = pb.ConfigAssociateGreedy()
config_greedy.forwardsBackwards = True
config_greedy.scoreRatioThreshold = 0.95
factory_association = pb.FactoryAssociate()
factory_association.set_score(pb.AssocScoreType.DEFAULT, feature_detector.get_descriptor_type())
associator = factory_association.greedy(config_greedy)

associator.set_source(desc0)
associator.set_destination(desc1)
matches = associator.associate()

print("Associated {} features".format(len(matches)))

# Visualize the images using a Java function
visualize_matches(image0, image1, locs0, locs1, associator.get_java_matches())

# TODO add support for python to java formatted matches

input("Press any key to exit")