International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 07 | July 2024
www.irjet.net
p-ISSN: 2395-0072
Signature verification using convolutional neural network (CNN) with Siamese model Zalak Chavda1, Stephy Patel2 1Student, Dept. of Computer Engineering (Software Engineering), Lok Jagruti Kendra University, Ahmedabad,
Gujarat, India
2Professor, Dept. of Computer Engineering, Lok Jagruti Kendra University, Ahmedabad, Gujarat, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In biometric authentication systems, verifying
CNNs are at tasks involving feature extraction and classification. Particularly Siamese networks have drawn interest because of their capacity to learn similarity metrics, which makes them appropriate for tasks involving pairwise comparisons like signature verification.
signatures is an essential task. In this paper, a Siamese Convolutional Neural Network (CNN) approach to signature verification on the CEDAR dataset is presented. The sigmoid function is used for classification, and Euclidean distance is used to measure similarity in the Siamese network. CNN layers make use of ReLU activation functions. The contrastive loss function and accuracy score are used to evaluate the model after the dataset has been divided into training and testing sets. Our test findings show how well the suggested approach works, obtaining a high degree of accuracy when separating real signatures from fakes.
3.METHODOLOGY 3.1 Dataset This study uses the CEDAR dataset, which includes multiple people's signatures that are both real and faked. The dataset is divided into testing and training sets to allow for a thorough assessment of the model's performance on untested data.
Key Words: Signature Verification, Convolutional Neural Network, Siamese Network, CEDAR Dataset, Euclidean Distance, Sigmoid Function, ReLU Activation
3.2 Preprocessing
1.INTRODUCTION
The signature images must be resized to a fixed dimension, grayscaled, and have their pixel values normalized as part of the preprocessing steps. By standardizing the input data, these procedures increase training process efficiency and boost model performance. The preprocessing function divides pixel values by 245 to normalize images and resizes them to a fixed size of 220x155 pixels. [4][2]
For many applications, including banking, legal documents, and access control, signature verification is essential to identity authentication. The accuracy and robustness needed for real-world applications are frequently lacking from traditional methods that rely on handcrafted features. Convolutional Neural Networks (CNNs) have demonstrated impressive gains in image recognition tasks, such as signature verification, since the introduction of deep learning. The application of a Siamese CNN architecture for signature verification is suggested by this study. The Siamese network, which is intended to learn similarity metrics between input pairs, is made up of two identical subnetworks that share weights. Training and assessment are conducted using the CEDAR dataset, a benchmark for signature verification. The sigmoid function is utilized to classify the data, and the Euclidean distance is used to measure the similarity between feature vectors. Nonlinearity is introduced within the CNN layers through the use of ReLU activation functions.
3.3 Convolutional Neural Network An instance of a deep learning system created especially for handling structured grid data, like pictures, is the convolutional neural network (CNN). Here are the main elements and characteristics : Convolutional Layers: In order to create feature maps, these layers apply a number of filters, also known as kernels, to the input picture. Specific patterns, such as edges, textures, or more intricate structures, are detected by each filter Activation Functions: Rectified Linear Unit (ReLU) activation functions are used to introduce non-linearity after convolutional operations, which aids in the network's ability to learn more intricate patterns.
2.RELATED WORK Traditional machine learning algorithms and the extraction of handcrafted features were the main strategies used in early approaches to signature verification. Recent developments in deep learning have shown how effective
© 2024, IRJET
|
Impact Factor value: 8.226
Pooling Layers: These layers use methods such as average or max pooling to minimize the spatial dimensions of the
|
ISO 9001:2008 Certified Journal
|
Page 264