MKLab researchers achieve first position in image forensics contest!
Researchers from MKLab are the winners of the first task of the Find-it! Fraud Detection Contest! The contest task called for approaches to detect documents containing at least one forgery in a flow of documents. The documents contained in the task dataset comprise 1969 scans of receipts.
The MKLab researchers set up a system based on image steganographic features, a bag of SVM classifiers and class-rebalancing during training. Using the training set provided by the organizers, this system managed to detect all forged documents without any error, achieving an F-score of 1.0.
More details on the results are available on the contest website.