- 期刊
An Improvement of the Interpretability for the Deep Learning Based Signature Examination Assistance Method
摘要
Automatic signature verification has been extensively researched for a long time and has already been used in many fields like banking, security, and other authentication purposes. However, human experts still play a dominant role in the field of forensic handwriting examination. Only a few studies have been conducted on the use of computers in assisting handwriting experts. There are also fewer attempts to perform the examination based on only a single known sample. We built a deep learning based assistance method for signature examination in our previous work. The method can deal with the problem of signature verification by single known sample, and is based on an explainable deep learning approach (by using deep convolutional neural network, DCNN). This paper is a continuation and refinement of our previous work. We refine the interpretability of the model and present application scenarios for assisting signature examination. After improving the interpretability of the model, the proposed method can be used as an assistant system by providing quantitative results. The visualized heatmaps can also be used to identify genuine or suspicious strokes in disputed signatures.