MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641075658 A) filed by Sr University on June 18, 2026, for A Robust Text Recognition System Using Masked Vision Transformer And Convolutional Decode.
Inventors include Ch. Aparna; and Dr. Rajchandar K.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: A ROBUST TEXT RECOGNITION SYSTEM USING MASKED VISION TRANSFORMER AND CONVOLUTIONAL DECODE The invention relates to a robust text recognition system capable of handling partially occluded, distorted, or noisy text images. The system integrates a Masked Vision Transformer (ViT) encoder, which divides input images into patches and masks a portion during training to learn contextual features, with a lightweight CNN decoder that refines encoded features and predicts character sequences. By incorporating simulated distortions during training, the system achieves strong generalization in real world scenarios. Postprocessing modules decode predictions into readable text, evaluated using accuracy, Word Error Rate (WER), and Character Error Rate (CER). The architecture achieves high accuracy and low error rates on benchmark datasets such as ICDAR 2015, while remaining lightweight and scalable. This invention provides a novel and efficient OCR framework suitable for applications in autonomous navigation, assistive technologies, and smart signage, offering improved tolerance to incomplete or degraded inputs compared to conventional systems.
Disclaimer: Curated by HT Syndication.
