MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641080243 A) filed by Koneru Lakshmaiah Education Foundation on June 30, 2026, for System And Method For Hybrid Multi-Stream Feature Fusion Using Handcrafted, Bag Of Features, And Deep Learning Representations For Robust Object Classification.
Inventors include Ms. M Srividya; and Dr. Venubabu Rachapudi.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The present invention discloses an intelligent computer vision framework for object classification using a hybrid multi-stream feature fusion architecture. The invention integrates classical Bag-of-Features descriptors, handcrafted visual descriptors, and deep convolutional neural network features through a unified Hybrid Feature Fusion Network (HybridFeatNet). Initially, object instances are extracted from annotated images using bounding-box localization. Multiple feature streams operate simultaneously, wherein one stream generates visual vocabularies using SIFT descriptors and K-Means clustering, another extracts Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP) descriptors, and a third stream generates semantic representations using a pretrained deep convolutional neural network. The extracted heterogeneous features are normalized, aligned, and fused into a unified feature vector before classification using a machine learning classifier. The proposed invention significantly improves classification accuracy, robustness against background clutter, and generalization across diverse object categories while maintaining computational efficiency suitable for surveillance, autonomous systems, industrial inspection, healthcare imaging, and edge AI devices.
Disclaimer: Curated by HT Syndication.
