Signature Verification Using a “Siamese” Time Delay Neural Network

Abstract

This paper describes the development of an algorithm for verification of signatures written on a touch-sensitive pad. The signature verification algorithm is based on an artificial neural network. The novel network presented here, called a “Siamese” time delay neural network, consists of two identical networks joined at their output. During training the network learns to measure the similarity between pairs of signatures. When used for verification, only one half of the Siamese network is evaluated. The output of this half network is the feature vector for the input signature. Verification consists of comparing this feature vector with a stored feature vector for the signer. Signatures closer than a chosen threshold to this stored representation are accepted, all other signatures are rejected as forgeries. System performance is illustrated with experiments performed in the laboratory.

Cross-modal adaptation for RGB-D detection

“…Multi-modal deep learning architectures have been explored previously in a generative context [36], [44], and parallel convnet architectures have been previously explored in the context of Siamese network learning [5], [7]. Given the ease of collecting annotations for an image classification task, as opposed to an object detection task, there have been many techniques proposed to train detectors from weak labels [41], [2], [1], [50].…”

“…However, ng and detection of large numbers of a key challenge problem in machine arkable progress has been made toect recognition, exploiting web-based luding ImageNet [1], PASCAL [2], [4]; recognition of thousands of catestrated in the ILSVRC challenge [1]. entation schemes are sometimes vittered real world conditions calls for s that perform multi-scale sub-window o detect a category of interest [5], [6]. odels (DPM) [5] are among the best n challenges that rigorously test den difficult conditions, e.g., PASCAL plementations with efficient inference are limited to models trained offline lly expensive training process and a s (e.g., the 20 PASCAL objects).…”

“…entation schemes are sometimes vittered real world conditions calls for s that perform multi-scale sub-window o detect a category of interest [5], [6]. odels (DPM) [5] are among the best n challenges that rigorously test den difficult conditions, e.g., PASCAL plementations with efficient inference are limited to models trained offline lly expensive training process and a s (e.g., the 20 PASCAL objects). At mbers of such a priori models could ll of ImageNet, or for typical search er, we show how to train and adapt deand on-demand, allowing the robotics perception system to the particular n. Unfortunately, models trained on large-scale datasets collected from the web often suffer in comparison to models trained from in-situ data in many domains [9].…”

A convolutional neural network approach for face verification

“…The concept of Siamese architecture in neural network was introduced by Bromley et.al. for signature verification [11]. This Siamese architecture consisted of two identical networks with same set of weights, and was combined at their outputs.…”

A Cross-domain and Cross-language Knowledge-based Representation of Text and its Meaning

“…Following the general Siamese neural network architecture (Bromley et al, 1993), Similarity Learning via Siamese Neural Network (S2Net) trains two identical neural networks concurrently. The S2Net receives as input parallel data with binary or real-valued similarity score and updates the model parameters accordingly .…”