VIS™:Visual Indexing and Search

VIS™ has a unique capability to derive, store and access metadata, content descriptors, and scene classification labels for large archives of images and video. VIS™ can help analysts to quickly retrieve image regions or video clips from large archives that are similar in activity, appearance, metadata (or combination of) to the query video clips or image examples.

VIS™ offers

  • Automated image and video labeling to generate annotations and context information for further intelligence.
  • Video activity search and retrieval.
  • Interactive query refinement for video and image content based search.
  • Flexible modules for advanced image registration.
  • Open architecture and customization.

View VIS™ in action

Automated image and video labeling

  • Automatically identify and locate the objects in the scene in fully automated manner.
  • Incorporates advanced image processing, computer vision, and multimedia indexing algorithms.
  • High accuracy (>85%) for most categories
  • System already tested with hundreds of gigabytes of data.
  • Rapidly provides actionable intelligence to commanders and war-fighters.

Joint object recognition and segmentation

  • Automatic detection and classification.
  • Accurate localization and estimation of object boundaries for better intelligence.

Interactive image/video search

  • Data-driven, not rule-driven description and interpretation.
  • Combination of supervised and unsupervised learning.
  • Improving accuracy with growing labels.
  • Retrieval from hundreds of gigabytes of data in seconds.
  • Background estimation and motion extraction from video frames.
  • Low- and high-level activity feature description.
  • Trajectory, Spatial, and Spatio-temporal descriptors.
  • Moving object classification (distinguish car vs. people).

View VIS™'s visual search on videos in action