Variability In Visual Small Color Difference Assessment: What It Means for Color Difference Formula Performance David Hinks, Renzo Shamey, Lina Cardenas, Seung Geol Lee, Rolf Kuehni, Warren Jasper North Carolina State University Raleigh, NC 27695 Introduction The majority of color-using industries require effective color difference prediction that accurately represents average perceptual assessments of the magnitude of color difference between two objects. Color difference formulas today are a critical tool for effective electronic color communication for color management in a product supply chain. Existing formulas are based on several different sets of perceptual data that have been established under various experimental conditions, using samples representing a diverse range of substrates and different groups of observers. In the textile industry, for example, the CMC (2:1) color difference formula is used as an international standard [1- 2]. In 2001, the International Commission on Illumination (CIE) recommended the CIEDE2000 formula, replacing the lower performing CIE94 [3]. Luo et al. reported accuracy of prediction for several formulas against average data from a large aggregate visual data set using four separate experiments. The aggregate data set was established as a fundamental part of the experiment that resulted in establishment of CIEDE2000. Using the PF/3 performance method, a value of 67.4 for the CIEDE2000 formula vs. 62.1 for CMC (2:1) was reported [4]. This paper reviews attempts to date to field test the performance of the latest CIE formula, proposes root cause variables to be investigated, and reports on a new set of visual color difference experiments currently underway. Field Testing of CIEDE2000 While the latest CIE formula is the highest performing equation when measured against the aggregate dataset, several independent experiments, each incorporating separate color difference pairs and visual observers, have not shown CIEDE2000 to perform statistically significantly better than CMC (2:1). In textiles, four unrelated field tests of CIEDE2000 vs. CMC (2:1) in three different countries resulted in a similar level of accuracy for the two formulas [5-8]. In the paints and coatings industry, initial work (e.g., by the Detroit Color Council) has also not shown conclusively that the new formula outperforms previous existing formulas [9]. Independent validation of any model is, of course, an essential first step to establishing confidence for its use in research fields or commercial applications. While the shortcomings of color difference equations in current industry standards are well known, industry-wide adoption of a major new protocol often takes many years to complete, and at considerable cost. To date, no data in the literature appears to provide a definitive answer to the disparity between the theoretical performance and the actual performance of CIEDE2000.