Forensic Verification of Handwritten Documents using Secure Multi Party Computation

The recent advance in the use of image processing techniques has benefitted many areas including forensic and digital verification in cybercrime detection. Image processing techniques are used for producing digital evidence in criminal activities including counterfeiting of currency notes, cheques, fake government documents, wills, official reports and certificates. In this article, we find the solution to help document verifier or examiner to find out and differentiate a fake document from the original document. However, we need to secure all the documents that are used for verification to maintain data security. Therefore, we use Secure Multi-Party Computation to preserve the documents securely. The expected solution focusses on image processing tools for verification of computer manipulated documents and capable of detecting altered written documents preserving the documents integrity all through the process. The objective of this article is to build an ImageNet model to recognize the fake document with an accuracy of greater than 97%, and to keep the false negative rate less than 5%. The accuracies achieved by ResNet50 and VGGNet16 using SMPC are 97.62% and 93.88% respectively in verifying the unauthorized document given by the author.