In this paper we propose a new device and illumination invariant image representation based on an existing greyscale image enhancement technique: histogram equalisation. Our method is based on the premise that the rank ordering of sensor responses is preserved across a change in imaging conditions (lighting or device). We set out the theoretical conditions under which this premise is true and we present empirical evidence which demonstrates that rank ordering is maintained in practice for a wide range of illuminants and imaging devices. We then show how we can exploit this rank invariance using histogram equalisation to derive an invariant image representation. Device and illuminant invariance are important in many imaging applications and in this paper we demonstrate the practical benefits of our new method in one such situation: the problem of image retrieval. We show that using the new invariant image representation to index into a database of images taken with a variety of device...
Graham D. Finlayson, Steven D. Hordley, Gerald Sch