Coin Counter Andres Uribe
what Find out the amount of money in a coin picture.
How – Classifier Build a coin classifier – Bayes Classifier – Nearest Neighbor – SVMs Segment coins and create feature vectors
How - DATA 80 images of standard US coins. 10 for each class: – Quarter: front and back – Dime: front and back – Nickel: front and back – Penny: front and back
How – Segmentation – Use of the Hough Transform to detect circles – Threshold selection to segment background
How – Features – Radial edge distribution: Detect edges in the coin image Construct a normalized edge radial histogram with 2, 4, 8, 16 and 32 bins.
Results – Classifier: Bayes: 72.5% SVM: yet to be implemented. NNR: yet to be implemented. – Money counter: First approach uses too much memory for the circle detection. Will use the best performing classifier.
Questions?