In an article published in the journal Scientific Data, researchers from Nanjing University of Science and Technology, China, proposed a novel scheme for handwriting identification and verification using Artificial Intelligence (AI) and texture features. This scheme aims to improve the accuracy and reliability of handwriting recognition and authentication in automation systems and intelligent process control.
Study: Spatial Variation-Dependent Verification for Enhanced Handwriting Identification using Artificial Intelligence. Image credit: kwarkot/Shutterstock
Background
Handwriting is a complex and dynamic biometric trait that reflects the individuality and personality of a writer. Handwriting verification is the process of determining whether a handwritten document or signature belongs to a claimed writer or not. It is widely used in various fields, such as security, forensics, banking, education, and healthcare.
However, verifying handwriting is a challenging task because of the variations in writing styles, writing objects, writing surfaces, and writing conditions. Moreover, handwritten signatures can be forged or imitated by malicious attackers, posing a threat to the integrity and confidentiality of the information.
Over the last decade, various techniques have been developed for handwriting verification, such as semantic analysis, feature extraction, machine learning, and deep learning. However, most of these approaches rely on subjective or manual features, which are not robust or scalable enough for real-world applications. Moreover, these methods do not consider the spatial and temporal variations of handwriting, which are crucial for capturing the subtle differences and similarities between writers.
About the Research
In the present study, the authors developed a novel scheme called Spatial variation-dependent verification (SVV) using textural features (TF) for handwriting identification and verification. TFs are the characteristics of the surface or region of an image that reflect its texture, such as smoothness, coarseness, contrast, etc., and can be used to distinguish different signatures based on their patterns and variations.
The SVV-based novel scheme uses a convolutional neural network (CNNs) to extract and match the TFs of handwriting, which are based on the pixel intensities and spatial variations of the handwritten signatures.
The novel scheme works as follows:
Identification point detection: First, it compares the pixel intensity of the handwritten and digital signatures to detect identification points, which are distinctive features of a signature that vary depending on its pattern, region, and texture.
TF extraction and spatial mapping: Second, it maps the identification points spatially with the digital signature to verify the matching of TFs. TFs are extracted among two consecutive identification points to avoid cumulative false positives, which are errors that occur when the system wrongly accepts a forged signature as genuine.
Signature verification: Third, it uses CNN to assist the verification process by generating new identification points and selecting the maximum matching features for varying intensity. This helps reduce the computation complexity and improve the accuracy of the verification.
The researchers used two datasets of handwritten signatures, one in English and one in Hebrew, and evaluated the performance of the scheme across five metrics: accuracy, precision, false positives, texture detection, and verification time. Furthermore, they compared the SVV scheme with three existing methods: Spatial Variability Simulated Neural Network (SV-SNN), Adversarial Variational Network (AVN), and Signature Verification based on Sound and Vibration for Handwriting Verification (SVSV).
Research Findings
The results show that the SVV scheme outperforms the other methods on all metrics, achieving an accuracy of 95.587%, a precision of 0.9644, a texture detection rate of 94.785%, a false positive rate of 0.0586, and a verification time of 0.831 seconds. The authors attribute the superior performance of the SVV scheme to its ability to capture the spatial variations and textural features of the signatures, as well as its use of a CNN to generate and select the optimal features for different intensities.
The developed strategy can be applied to various domains that require handwriting identification and verification, such as banking, e-commerce, education, healthcare, law enforcement, security, forensics, etc. This can provide a secure, reliable, and efficient way to authenticate and authorize digital or handwritten signatures, as well as to detect and prevent forgery, identity theft, and fraud. Furthermore, this can be extended to other images with textural features, such as fingerprints, faces, iris, etc.
Limitations and Conclusion
The authors acknowledge the following limitations in their study:
- It relies on the availability and quality of digital signatures, which may not always be accessible or consistent in real-world scenarios.
- It does not consider the temporal aspects of handwriting, such as speed, pressure, and rhythm, which may also affect the identification and verification process.
- The used datasets are limited to English and Hebrew languages and may not generalize well to other languages or scripts.
- The feature extraction methods are based on existing techniques and may not capture all the relevant aspects of handwriting styles and gender differences.
In summary, the authors proposed a novel scheme for handwriting identification and verification using artificial intelligence-assisted textural features. This scheme was more accurate, precise, and efficient than the existing methods and could handle different handwriting styles and genders.
LG's EXAONE 3.5 Sets New Standards in Generative AI