TY - JOUR
T1 - Learning the micro deformations by max-pooling for offline signature verification
AU - Zheng, Yuchen
AU - Iwana, Brian Kenji
AU - Malik, Muhammad Imran
AU - Ahmed, Sheraz
AU - Oyama, Wataru
AU - Uchida, Seiichi
N1 - Funding Information:
This work was supported by JSPS KAKENHI (Grant Number JP18K11373 and JP17H06100), China Scholarship Council (Grant Number 201706330078 ), Innovation and Cultivation Project for Youth Talents of Shihezi University (Grant Number CXPY201905), and Financial and Science Technology Plan Project of Xinjiang Production and Construction Corps (Grant Number 2017DB005).
Publisher Copyright:
© 2021 Elsevier Ltd
PY - 2021/10
Y1 - 2021/10
N2 - For signature verification systems, micro deformations can be defined as the small differences in the same strokes of signatures or special writing habits of different signers. These micro deformations can reveal the core distinction between the genuine signatures and skilled forgeries. In this paper, we prove that Convolutional Neural Networks (CNNs) have the potential to extract those micro deformations by max-pooling. More specifically, the micro deformations can be determined by watching the location coordinates of the maximum values in pooling windows of max-pooling. Extensive analysis and experiments demonstrate that it is possible to achieve state-of-the-art performance by using this location information as a new feature for capturing micro deformations, along with convolutional features. The proposed method outperforms the state-of-the-art systems on four publicly available datasets of different languages, i.e., English (GPDSsynthetic, CEDAR), Persian (UTSig), and Hindi (BHSig260).
AB - For signature verification systems, micro deformations can be defined as the small differences in the same strokes of signatures or special writing habits of different signers. These micro deformations can reveal the core distinction between the genuine signatures and skilled forgeries. In this paper, we prove that Convolutional Neural Networks (CNNs) have the potential to extract those micro deformations by max-pooling. More specifically, the micro deformations can be determined by watching the location coordinates of the maximum values in pooling windows of max-pooling. Extensive analysis and experiments demonstrate that it is possible to achieve state-of-the-art performance by using this location information as a new feature for capturing micro deformations, along with convolutional features. The proposed method outperforms the state-of-the-art systems on four publicly available datasets of different languages, i.e., English (GPDSsynthetic, CEDAR), Persian (UTSig), and Hindi (BHSig260).
UR - http://www.scopus.com/inward/record.url?scp=85106320679&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85106320679&partnerID=8YFLogxK
U2 - 10.1016/j.patcog.2021.108008
DO - 10.1016/j.patcog.2021.108008
M3 - Article
AN - SCOPUS:85106320679
VL - 118
JO - Pattern Recognition
JF - Pattern Recognition
SN - 0031-3203
M1 - 108008
ER -