Objective ink color comparison through image processing and machine learning

Making changes or additions to written entries in a document can be profitable and illegal at the same time. A simple univariate approach is first used in this paper to quantify the evidential value in color measurements for inks on a document coming from a different or the same source. Graphic, qualitative discrimination is then obtained independently by applying color deconvolution image processing to document images, with parameters optionally optimized by support vector machines (SVM), a machine learning method. Discrimination based on qualitative results from image processing is finally compared to the quantitative results of the statistical approach. As color differences increase, optimized color deconvolution achieves qualitative discrimination when the statistical approach indicates evidence for the different source hypothesis.