Vocabulary Tree Hypotheses and Co-Occurrences

Proceedings of the 12th Computer Vision Winter Workshop (CVWW'07) - BEST PAPER AWARD

This paper introduces an efficient method to substantially
increase the recognition performance of a vocabulary tree
based recognition system. We propose to combine the hypothesis
obtained by a standard inverse object voting algorithm
with reliable descriptor co-occurrences. The algorithm
operates on different depths of a standard k-means
tree, coevally benefiting from the advantages of different levels
of information abstraction. The visual vocabulary tree
shows good results when a large number of distinctive descriptors
form a large visual vocabulary. Co-occurrences
perform well even on a coarse object representation with a
few number of visual words. We demonstrate the achieved
performance increase, robustness to occlusions and background
clutter in a challenging object recognition task on a
subset of the Amsterdam Library of Object Images (ALOI).