ImagEval task 1: Retrieving transformed images with new local descriptors based on dissociated dipoles

This paper describes new local photometric descriptors based on dissociated dipoles for transformed images or rigid objects retrieval. Dissociated dipoles are non local differential operators proposed by Balas et al. [1] which have been proved to be more stable than purely local standard differential operators. In this paper, we define and compute specific oriented dissociated dipoles around multi-resolution color Harris points and we form 20-dimensional normalized features, invariant to rotation, affine luminance transformations, negative or flip. In a comparison with extensively used SIFT descriptors, we show that such descriptors are more efficient while containing 6 times less information. This allows the complete retrieval to be both more efficient and faster. This strategy ranked first in ImagEval benchmark, which, as far as we know, is the only competition including a transformed image recognition task (or content-based copy retrieval task).