Semantic Source Coding for Flexible Lossy Image Compression

摘要

Semantic Source Coding for Lossy Video Compression investigates methods for Mission-oriented lossy image compression, by developing methods to use different compression levels for different portions of an image based on their utility in understanding the scene depicted. We have used semantic methods, including pattern discovery, wavelet-based segmentation and texture segmentation to extract the essential predictive information in the image. Described are mission oriented approaches to image pre-processing, image and video segmentation, quality comparison, compression, and two complete systems for lossy video compression.