Organizing and indexing media
We are developing several media understanding technologies to help consumers organize their collections of images and other media. These include:
- face detection
- image quality assessment
- scene type classification
- video segmentation
- video keyframe extraction
- time-based clustering of photos and video
This work builds on HP's in-depth understanding of how consumers use and interact with their photos, videos and other media, as well as on HP's existing portfolio of tools for analyzing, enhancing, capturing and printing media.
We have a variety of projects aimed at making it easier for consumers to create captivating albums, photo books, slideshows and other presentations.
One project looks at scalable storytelling, which combines time-based clustering with image quality assessment (or ‘image triage') to automatically generate slideshows of any length from large collections of personal photos.
For print-based presentations, researchers developed Blocked Recursive Image Composition (BRIC) layout engine that automatically sizes and places photos on a page to make the most effective use of the available space without requiring the photos to be cropped.
HP Labs' SmartFrame technology automatically analyzes photos and creates a color matting for each photo that complements the colors in the photo.
We have also developed algorithms that analyze image content to automatically crop sets of photos, allowing busy users to make better-looking albums.
For a more informal, collage-type layout, researchers developed the Wacky Layout algorithm, which places photos at different orientations and allows them to overlap. Wacky Layout analyzes each photo to ensure that the most important parts of the image are visible and the overlaps generally occlude only the unimportant areas.
For dynamic presentations on televisions or computer screens, HP Labs is developing an automatic slideshow-creation engine that uses cinematic principles from filmmaking to create compelling shows where the consumer's photos are the 'actors.'