An energy spreading transform approach to efficient image retrieval

When searching a large database of images for ones that are visually similar to a particular reference images, it is desirable to have efficient means for making coarse comparisons between images. This paper describes an approach for efficient image retrieval using an energy spreading Hadamard transform arising from an orthogonal basis of unimodular sequences called the Prometheus orthonormal set (PONS). The energy spreading properties of this PONS transform suggest that global comparison of key features of images can be accomplished using any subset of the PONS transform coefficients.