The Ineffectiveness of the Correlation Coefficient for Image Comparisons

: Pearson's r linear correlation coefficient is widely used for comparing images. Image processing experts appear to be cognizant of its serious limitations. This information, however, has not been communicated well to non-experts (including those working on security applications), who sometimes use the correlation coefficient without being aware of its problems. We discuss the disadvantages of the correlation coefficient and show specific examples of strikingly poor performance in the context of security. KEYWORDS: correlation coefficient, tamper detection, tags, seals, image comparison Introduction Pearson's correlation coefficient, r, is widely used in statistical analysis, pattern recognition, and image processing [1-8]. Applications for the latter include comparing two images for the purposes of image registration, object recognition, and disparity measurement. For monochrome digital images, the Pearson correlation coefficient is defined as [1,4,7]: r = (x i - x m )(y i - y..