MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124066 A) filed by Koneru Lakshmaiah Education Foundation on October 16, 2026, for Explainable Ai-Driven Hybrid Cnn-Transformer Framework For Robust And Real-Time Facial Emotion Recognition.
Inventors include Thota Naga Sri Satya; Boppana Vijay Chowdary; and Talasila Taraka Rajeswar Rao.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A real-time facial emotion recognition system (100) detects and tracks faces in camera or video frames (106), aligns and pre-processes each face (108), and identifies occluded facial regions (118). A hybrid model (110) extracts a 6 x 6 feature map with residual convolution blocks, converts it into tokens with a class token, and processes them with a transformer encoder whose self-attention is masked for occluded tokens, giving probabilities of seven emotions. The probabilities are smoothed per tracked face over frames (114); the emotion is output with its confidence when it reaches a calibrated threshold, otherwise an uncertain result is output. An explainability module (116) fuses attention rollout and gradient-weighted class activation maps into a heat map, computes the contribution of each facial region, and flags predictions whose dominant region is occluded. The system runs on computers and edge devices.
Disclaimer: Curated by HT Syndication.
