This repository contains pre-trained signature verification models on three datasets: BhSig260Hindi, BhSig260Bengali, and Cedar.
The BhSig260Hindi dataset consists of 260 Hindi signatures collected from different individuals. It is primarily used for Hindi signature verification tasks. Each signature in the dataset is labeled with the corresponding identity of the individual who provided the signature. The dataset captures the variability present in Hindi signatures, including different writing styles, stroke patterns, and sizes. It is a valuable resource for training and evaluating signature verification models specifically designed for Hindi signatures.
The BhSig260Bengali dataset comprises 260 Bengali signatures obtained from various individuals. It is specifically designed for Bengali signature verification tasks. Similar to the BhSig260Hindi dataset, each signature in this dataset is associated with the identity of the individual who provided the signature. The dataset encompasses a range of Bengali signature characteristics, such as different writing speeds, pen pressure variations, and overall style differences. It serves as a valuable benchmark for evaluating the performance of signature verification models on Bengali signatures.
The Cedar dataset is a comprehensive collection of signatures captured from multiple writers. It serves as a benchmark for signature verification research and development. The dataset includes signatures from various individuals, capturing a wide range of writing styles, patterns, and characteristics. It is designed to be a challenging dataset for signature verification tasks, with variations in signatures caused by different writing utensils, writing conditions, and writing habits. The Cedar dataset is widely used in the signature verification community to evaluate and compare the performance of different models and algorithms.
The repository includes the following trained models:
Signature verification model based on SCNN architecture.
Signature verification model based on the Siamese network architecture using the SigNet model.
Signature verification model based on the ResNet-50 architecture.
Signature verification model based on the Siamese network architecture using the ResNet-50 model.
Signature verification model based on the Siamese network architecture using the ResNet-18 model.
Signature verification model based on the Siamese network architecture using the EfficientNet-B0 model.
These models are trained to perform signature verification tasks, which involve determining the authenticity of a given signature. The models are capable of differentiating between real and forged images of signatures, providing a mechanism to detect fraudulent or unauthorized signatures.
To use these pre-trained models, follow the instructions below:
- Clone the repository:
https://github.com/dhruvkej9/SignatureVerification.git- Install the necessary dependencies (if any) specified in the import of each file.
Contributions to this repository are welcome! If you have trained additional signature verification models on the mentioned datasets or have any improvements to existing models, feel free to submit a pull request.