A Robust Visual Identifier Using the Trace Transform

This paper introduces an efficient image identifica tion method designed to be robust to various image modif ications such as scaling, rotation, compression, flip and gr ey scale conversion. Our method uses trace transform to ext ract a 1D representation of an image, from which a binary str ing is extracted using a Fourier transform. Multiple compo nent descriptors are extracted and combined to boost the robustness of the identifier.. Experimental evaluat ion was carried out on a set of over 60,000 unique images a nd one billion image pairs. Results show detection rate of over 92% at false-positive rate below 1 per million, with ma tching speed exceeding 4 million images per second.