Paper Title:
Salient Video Stills: Content and Context Preserved
Author(s):
Laura Teodosio and Walter Bender
Main Point(s):
The paper proposes a new class of images
called salient stills that reflect the aggregate of temporal changes
that occur in a sequence of images with the salient features
preserved. The motivation of using salient stills is to reduce
redundancy in a sequence of frames. The salient still contains
multi-resolution patches, a larger field of view, or higher overall
resolution than any individual frame in the original image sequence.
To construct such a still image, the optical flow between successive
frame pairs is first computed, then affine transformations are
applied, which translate, scale and warp each frame into a single,
high-resolution raster, finally a weighted temporal median filter is
applied to the high-resolution image data to get the result
image. Some enhancements and extensions are also discussed.
Possible Use(s):
Salient stills can be used to create panoramic pictures. They are also
good for icons in a video database that served for content-based retrieval
purpose. One more use is to monitoring an important place. as a "video radar".
Extensions:
Extended to modeling moving objects with salient features.
Jing Huang's CS631 course page