MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641093752 A) filed by Santhiram Engineering College on August 03, 2026, for An Efficient Machine Learning Framework For Handwritten Signature Authentication And Validation.
Inventors include Syed Sadia Fatima; Dr. Sowmya Gali; Shaik Farhath; Mancha Subhash; Pall Vignesh; Manchala Koti Bharath Kumar Reddy; and Shaik Fayaz Ahammed.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: Digital and web-based systems have become increasingly important, leading to a growing need for secure and dependable identity authentication methods. Handwritten signatures are still widely used in the legal, financial, and institutional sectors, and thus have a long history as acceptable forms of identification. The challenge is the variability among signatures created by the same person or user (i.e. intra-user variability), the potential for a highly skilled person to forge a signature, and differences in the environment in which a signature was created. We propose a system for verifying handwritten signatures that will utilize convolutional neural networks (CNNs) to learn distinguishing features of handwriting from images of the signatures. Our system uses a strong preprocessing approach that includes converting the signature images to grayscale, reducing noise from the images,using adaptive thresholding, and extracting, regions-of-interest (ROI). A deep learning model will be built from the signature images and will capture the underlying complexity of the signature's structure and strokes in order to eliminate the need for manually created features. In order to enhance the verification system's robustness, the multi-sample enrollment strategy utilizes multiple genuine signatures to create a model with a greater range of handwritten variation. Similarity checks between enrolled signature samples and test signature samples utilize adaptive decision, thresholds during the verification process. This system differentiates genuine signatures from forged signatures by utilizing these adaptive decision thresholds during the verification phase. The use of Python, OpenCV, and a Flask web framework for developing and deploying the verification system allows for rapid deployment and use in practical applications. Experimental evaluations of the proposed approach show that high verification accurate levels are achieved with low levels of false rejects and false accepts. This provides an indication that deep learning methods feature extraction provide significantly higher levels of verification accuracy and reliability than traditional rules- based methods or handcrafted feature extraction methods, thus providing a viable solution for secure biometric authentication in real-world applications.
Disclaimer: Curated by HT Syndication.
