Intelligent Automation, Inc. — Department of Defense SBIR Phase I: N192-097

Description

To address the need for advanced video compression capability, IAI proposes a robust solution which works across many different video formats, by first transcribing the video to a set of color jpeg images at a specified frame rate and dimensions. The resulting images are then compressed by a parallel multi-stage scheme, which takes advantage of nearby image channel correlation, approximate linear dependence under transformation, and uses image tailored lossless compression methods. In particular, we will utilize an algorithm, which takes advantage of high correlation between signals in a cluster, in order to encode the signals efficiently in terms of a single pillar member of the cluster. We will also apply sparsity preserving low rank matrix factorization techniques, such as the interpolative decomposition, to take advantage of approximate linear dependence between RGB channel data in nearby images. For lossless compression, we will consider a setup using inversion coding, based on distances between occurrences of repeating symbols, as well as recent work on hierarchical predictions. Supplementing this will be a parallel module for approximate decompression and video playback.