Offline signature verification using DAG-CNN

Abstract

This paper presents the implementation of a DAG-CNN which aims to classify and verify the authenticity of the offline signatures of 3 users, using the writer-independent method. In order to develop this work, 2 databases (training / validation and testing) were built manually, i.e. the manual collection of the signatures of the 3 users as well as forged signatures made by people not belonging to the base and altered by the same users were done, and signatures of another 115 people were used to create the category of non-members. Once the network is trained, its validation and subsequent testing is performed, obtaining overall accuracies of 99.4% and 99.3%, respectively, showing the features learned by the network and verifying the ability of this configuration of neural network to be used in applications for identification and verification of offline signatures.

Comparison of feature extraction and normalization methods for speaker recognition using grid-audiovisual database

“…(3) where: l is the channel incident, m represents the frame incident, L is the number of frequency channels, and is the forgetting factor which is equal to 0.999, whereas the T[m,l] represents the time-frequency normalization. The power-law nonlinearity is produced by a value of 1/15 which is empirically chosen to give acceptable accuracy in white noise and without any significant impact on recognition accuracy in clean speech, as shown in (4) [19]: (4) where U[m,l]: is the normalized power. Table 1 summarizes the difference between the MFCCs and PNCCs features.…”

“…Biometrics characteristics can be divided into two main types [1]: Behavioural and Physiological traits. Examples of the behavioral traits include Voice, Signature, Gait, and Keystroke [2][3][4][5], and [6]. On the other hand, Physiological traits include Iris, Retina, Face, Ear, DNA, Hand Geometry, Palm, and Fingerprint [7][8][9][10][11][12].…”

Assistant robot through deep learning

“…The kinematic model of the arm is required to be able to establish the displacement of this within its workspace, in which both the tools and the user's hand are located. In Figure 5, it can be observed the geometric model that allows inferring (5) to (14), through which it can be set the angular movements of the robot. From the top view, the angle of joint 1 ( ) and the X component of point P of the final effector are observed, through which (5) and (6) are established.…”

“…The angle of joint 2 is calculated using β and α using (11) and (13). Since the angle of the second joint can have two different values, depending on the grip of the manipulator using elbow up or elbow down, the (14) is used.…”

“…CNNs have shown high performance in the recognition of objects [11] and today several architectures of this type of network are applied in machine vision applications [12]. Some examples of CNN application are framed in fields such as text classification, with web applications [13] or signature recognition [14], thus showing the versatility of these networks in pattern recognition. However, variations of CNN base architectures are still being developed in order to improve aspects of learning, such as the case of variations in the depth at which the objects are, as shown in [15,16].…”