Can texture and image content retrieval methods match human perception?

Texture is widely used in content based image retrieval (CBIR), and there have been a number of studies over the years to establish which features are perceptually significant. However, it is still difficult to retrieve reliably images that the human user would agree are "similar". Different studies tend to use different image databases, but the Brodatz textures are widely recognised, and form a standard test set. This paper outlines a number of performance measures, and uses the results from a human study to classify textures from the Brodatz textures dataset to define "relevance" and hence compute correlation between each technique and the perceptually derived ranking. Our results how that in general image content-based retrieval methods do not match human perception well.