Colour Space Selection in Image Hashing: An Experimental Study

Abstract

ABSTRACTColour space has been widely used in digital image processing, but its selection is rarely discussed in image hashing. Aiming at this problem, we discuss colour space selection by evaluating classification performances of typical hashing algorithms under YCbCr colour space, CIE L*a*b* colour space, HSV colour space, and HSI colour space. Our contributions are two sides. (1) We find that the regularly used YCbCr colour space cannot reach desirable classification performance and HSV colour space outperforms other colour spaces. (2) We analyse classification performances of D-DCT hashing, NMF-NMF-SQ hashing, RT-DCT hashing, and GF-LVQ hashing under different colour spaces, which are the first reports of these algorithms. Receiver operating characteristic graph is used to analyse classification experiments with large data-sets of 2220 similar image pairs and 19,900 different image pairs.