Signature Recognition and Forgery Detection Using Deep Learning.
Signature verification is being applied broadly in industries. Characteristics like psychological and behavioural can be used to verify and authenticate an individual, these are called biometrics. Aim: This research focus is on offline signature recognition and forgery detection using Convolutional Neural Network (CNN) and Siamese Neural Network. Method: This study made use of the CEDAR (Center of Excellence for Document Analysis and Recognition) data set, which has a combined total of 1,320 authentic signatures and 1,320 forgeries. It maintains a balance between false rejection rate (FRR) and false acceptance rate (FAR). The model was trained and evaluated on a comprehensive dataset, resulting in an FRR of 0.0161 and a FAR of 0.2903. A comparison of the model's performance revealed that this approach performed favourably in terms of FRR and FAR. These results suggest that this study’s approach is a promising solution for signature recognition and forgery detection applications.
Download Paper