Improvement of histogram-based image retrieval and classification

Histogram-based techniques are commonly used in image retrieval ranging from basic color histograms to sophisticated histograms of various local feature vectors. Our approach considers multidimensional histograms of locally invariant features. In order to get rid of the discontinuous behavior of traditional histograms we develop a fuzzy histogram which does no sharp assignment of a feature vector to one bin only but performs a weighted assignment to all neighboring bins. This improvement is not restricted to the kind of feature histograms considered here but could generally improve histogram-based retrieval or classification techniques.