Perception Based Image Retrieval

We created an image indexing system based on some of the known properties of the early stages of human vision. We used a color space known to underlie the second stage of human color vision and stored chromaticity and luminance information in two logarithmic-radial histograms. A third, spatial index encodes – in analogy to the spatial frequency representation in the visual cortex – information about orientation and spatial scale. The indices were evaluated by comparing the computed similarity values with human judgments quantitatively and objectively in a 2AFC design. For the experiments we used a heterogeneous database of 60,000 digitized photographs.