MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641075659 A) filed by Sr University on June 18, 2026, for A Deep Learning–based System For Authenticity Verification Of Human Faces.
Inventors include Goolla Mamatha; and Rajchandar K.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: A DEEP LEARNING–BASED SYSTEM FOR AUTHENTICITY VERIFICATION OF HUMAN FACES A computer-implemented text recognition system is disclosed. The system comprises a Masked Vision Transformer (ViT) encoder configured to divide input images into patches and mask a portion of the patches during training to learn contextual features. The ViT encoder utilizes contextual attention to reconstruct missing or degraded portions of text. The system further comprises a convolutional neural network (CNN) decoder configured to refine encoded features and predict character sequences for recognizing partially occluded, distorted, or noisy text. Input images are preprocessed through resizing and normalization prior to patch division, while postprocessing decodes predicted features into readable text sequences. Training incorporates simulated distortions including blur, noise, and occlusion to improve generalization under real-world conditions. The CNN decoder refines temporal features without requiring recurrent neural networks (RNNs) or long short-term memory (LSTM) layers. The architecture provides improved efficiency, scalability, and tolerance to incomplete inputs, enabling accurate text recognition with low error rates and suitability for deployment in resource-constrained environments and applications including autonomous navigation, assistive technologies, and smart signage.
Disclaimer: Curated by HT Syndication.
