This paper discusses techniques for visualizing structure in video data and other data sets that represent time snapshots of physical phenomena. Individual frames of a movie are treated as vectors and projected onto a low-dimensional subspace spanned by the principal components. Movies can be compared and their differences visualized by analyzing the nature of the subspace and the projections of multiple movies onto the same subspace. The approach is demonstrated on an application in neurobiology in which the electrical response of a visual cortex to optical stimulation is imaged onto a high-speed photodiode array to produce a cortical movie. Techniques for sampling movies over a single trial and multiple trials are discussed. The approach provides the traditional benefits of principal component analysis (compression, noise reduction and classification) and also allows visual separation of spatial and temporal behavior.
CR Categories and Subject Descriptors: I.3.3 [Computer Graphics]: Picture/Image Generation - Viewing Algorithms.
Additional Keywords: Scientific visualization, animation, video analysis.
This paper appears in IEEE Visualization '98 pages 217-224 and is copyrighted by IEEE. The material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therin are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints of the copyright. In most cases, these works may not be reposted without explicit permission of the copyright holder.