Creating efficient codebooks for visual recognition | IEEE Conference Publication | IEEE Xplore