Compression and segmentation of images using an intersubband wavelet probability model
Study of the statistical qualities of images can lead to elegant solutions for many classical image processing and computer vision problems. Image compression, restoration, interpolation, and texture segmentation can be written in terms of probabilistic decision making, where the prior model is the set of “natural images.” Recently, we introduced a statistical characterization of natural images in the wavelet transform domain. This characterization describes the joint statistics between pairs of subband coefficient magnitudes at adjacent spatial locations, orientations, and scales. This paper describes our work in image compression and unsupervised segmentation that utilizes inter-subband dependencies. In order to support unsupervised segmentation with a large feature set, this paper presents a novel unsupervised algorithm to reduce feature dimensionality called “Unsupervised Linear Discriminant Analysis” (ULDA) based upon Fisher's Linear Discriminant Analysis.
Computer science|Electrical engineering
Buccigrossi, Robert William, "Compression and segmentation of images using an intersubband wavelet probability model" (1999). Dissertations available from ProQuest. AAI9937704.