Comparative Study of Faster Region-Based Convolutional Neural Networks with Inception V2 and Single Shot Detector with Inception V2 on Their Signature Detection Capabilities

Abstract

AbstractSignature detection and verification in offline handwritten signature printed document is a problem when solved helps to target forgery and reduce man hours. With challenges like stray marks, several methods were proposed and eventually with the rise of vision recognition methods and Deep Learning, application program interfaces have been developed to detect signatures. Using TensorFlow APIs, images that are converted to radio frequency (RF) format are processed to detect signatures. This paper intends to evaluate two different object detection algorithms, faster region-based convolutional neural networks and single shot detector with inception V2, on signature dataset. Various standard metrics such as train and test time, loss rate and precision at area under curve etc. are used to compare their performance and fitment for the problem statement.KeywordsSignature detectionObject detectionFaster region-based convolutional neural networks (faster RCNN)Single shot detector (SSD)Inception V2