Color Features for Image Fingerprinting Marios A. Gavrielides, Elena Sikudova, Dimitris Spachos, and Ioannis Pitas AIIA Laboratory, Computer Vision and Image Processing Group, Dept. of Informatics, Aristotle University of Thessaloniki, Box 451 Thessaloniki, GR-54124, Greece Abstract. Image fingerprinting systems aim to extract unique and robust image descriptors (in analogy to human fingerprints). They search for images that are not only perceptually similar but replicas of an image generated through mild im- age processing operations. In this paper, we examine the use of color descriptors based on a 24-color quantized palette for image fingerprinting. Comparisons are provided between different similarity measures methods as well as regarding the use of color-only and spatial chromatic histograms. 1 Introduction Image fingerprinting (or perceptual hashing) refers to the extraction of a unique descrip- tion of an image that would be resilient to transformations, in an analogous manner to human fingerprints. Detecting transformed versions of images could be used to fight piracy of such material. Image fingerprinting differs from image watermarking in the sense that watermark- ing involves embedding information into the image, whereas fingerprinting, as defined here, involves descriptors extracted from the image content. Fingerprinting can be used to search for those copies of an image that have already been circulating in the internet with no watermarks embedded on them. In order for such a system to be successful, it has to be robust against a number of frequent attacks, have good discriminating ability to avoid the retrieval of false alarms, provide efficient storage of extracted image descriptors that would be used for image matching and an efficient search mechanism that would compare the image description of the query image to those in a database of image descriptors. A number of approaches [1], [2], [3] have been presented in the literature. In this paper, we examine the use of color-based descriptors for an image finger- printing system and its robustness to most common attacks. We demonstrate the effect of various color descriptors including color-only and color-spatial information, reduced number of colors, and different types of histogram-similarity measures. 2 Color-Based Fingerprint Extraction The fingerprint extraction procedure involves the quantization of the image colors and the calculation of color histograms based on the resulting colors. We used a quantiza- tion method based on a pre-defined color palette known as the Gretag Macbeth Color G. Antoniou et al. (Eds.): SETN 2006, LNAI 3955, pp. 494–497, 2006. c  Springer-Verlag Berlin Heidelberg 2006