Pre-processing of Signature Images for Detection and Verification

The companies receive a lot of scanned or photographed documents from their customers. Therefore, an important technology is to automatically detect the signatures on the documents and then match them with the customer's signature in the system for verification. However, since overlapping signature and stamp makes it harder to solve the problem, some pre-processing operations are necessary. The object detection models that are used for detecting the signatures on the documents could not perform well when they are trained with clean signature images. To address this problem, in this study a new dataset is generated by combining some fake stamps with clean signatures to train the model with them. In the next step, the detected stamped signatures are matched with the clean reference signatures. However, due to stamps, this process cannot be performed successfully. For this reason, in order to remove the stamp, an image processing system is developed. Obtained results show the capability of image processing techniques for the challenging problem of cleaning the stamped signature image.