Image and Video Indexing and Search
SRI offers advanced computer vision capabilities to efficiently and automatically categorize and search large media datasets.
Currently, image and video search is mostly based on tags derived from text associated with the images or specified manually. Manually tagging and organizing large volumes of digital media can take hundreds of hours, and important details can be missed. Similarly, associated text is often ambiguous or missing. One of SRI’s latest advancements in machine learning technology is the ability to efficiently and automatically tag, categorize and index large volumes of images and videos using solely the imagery content.
Building on capabilities in visual indexing, feature extraction, automated tag extraction, deep learning and clustering, SRI’s technology enables true content-based search and retrieval. It can quickly explore and index images and videos using both semantic and visual features and attributes, including locations, logos, objects, shapes, colors, entities, actions and events.
The technology creates non-semantic metadata (a number versus a pre-loaded word) that represents visual features in a photo or video. This allows the search engine to recognize concepts it has not been trained to recognize. It can be used in a variety of scenarios to sort images by visual similarity—from personal photo libraries to real estate listings. In video, it recognizes and allows users to search dynamic content and concepts. For example, users can search for where the cake was cut in a birthday party video.
Automated indexing and search technology unlocks the visual intelligence contained in large media datasets, providing an effective way to organize and search visual content.
SRI is developing a system for the semi-automated geolocalization of metadata-free images and videos to find a location of interest.
SRI is developing a novel search technology to quickly find events of interest in very large video collections.