TITRE

1. TITRE Parametric colour transformation: (a) genuine and (b) fake samples. (a) (b) Classification capability of individual (blue line) features and their gradual combination (brown line). Holes in character images: (a) a genuine and (b) a fraudulent samples Histogram of hue of character strokes: (a) genuine and (b) fake sample Figure 1. Role of dominant intensity. Authentication of Currency Notes through Printing Technique Verification Ankush Roy1, Biswajit Halder2, Utpal Garain3 1 Student, Dept. of Elect. Engg Jadavpur University,Kolkata,India 2 Mallabhum Institute of Technology, Bishnupur, WB, India 3 Indian Statistical Institute, Kolkata,India Introduction • Average Color • A colour transformation is made on two indivisualcolour streams to produce • a new colour matrix • A new method to detect counterfeit paper currency • The method is based on verifying the printing process used (Intaglio printing) as a security feature • Twoclassifiers are used: Neural Network & SVM • The discriminatory power of the featuresisshownusing LDA • This algorithm is evaluated on real data S(i) = pBblue (i) + (1 – p)Bblack(i) , 0<p<0.5 Features • Dominant intensity • Perimeter Based Edge Roughness: • pa is the perimeter of the actual image • pb is perimeter of the filtered binary image • EPBER is the perimeter based edge roughness • Area difference: • Area difference = │(Aotsu+sc – Aotsu ) │/Aotsu • A is the binarised image with a normalised parameter (sc) Figure (c) and (d) show the masked images of two character images extracted from two currency notes (one genuine and one fake respectively). Figures (e) and (f) show the histograms of gray levels as computed on the masked images. Experimentation • Hole Count Clustering of Currency Note Printing Techniques using K-Means Holes in character images: (a) a genuine and (b) a fraudulent sample • Average Hue (b) (a) • Contrast • RMS contrastisused to tap the slightdifference • in brightness (or glossiness) that the humaneyefails to • recognise • Keytone • The mean gray value of all the pixels. The value of key tone indicates whether • the bulk of information in an image is stored in the high/middle/low intensity zone Classification of Currency Note Printing Techniques Using SVM • Correlation Coefficient: • A is the grayscale image and B the • corresponding binary image after • adding a normalized parameter (sc/255) • over the Otsu limit NN- based Classification too achieves very high accuracy (about 99.5%, 0.5% error is attributed to true negative) ICVGIP 2010 contact : utpal@isical.ac.in Conseil Scientifique, 4 Mai 2007