Browse Prior Art Database

Lossless, Grey-Scale Image Compression, Based on Reduced Parameter Model and Coder

IP.com Disclosure Number: IPCOM000038966D
Original Publication Date: 1987-Mar-01
Included in the Prior Art Database: 2005-Feb-01
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Arps, RB: AUTHOR [+2]

Abstract

A method is described for reducing the number of degrees of freedom in a Markov source model for the purpose of image compression, comprising the steps of: (1) mapping the k-component Markov states into a smaller set of single-component states by a linear collapsing function, and (2) reducing each conditional probability distribution to a few parameters by matching them to known functions with the best fit. The Markov process is a basis of a general model with memory for compressing image sources. The image is viewed as a condition pixel source (X/S) producing a time sequence of N-bit pixels: XX{xn=ne{0,1,.....(2N-1)}} A kth-order Markov source model is defined by the conditional probabilities {P(X/S)} and stationary probabilities {(S)}, where S is a Markov state consisting of k previous pels (X1,.X2,.Xj,.Xk), 1&j&k.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 53% of the total text.

Page 1 of 1

Lossless, Grey-Scale Image Compression, Based on Reduced Parameter Model and Coder

A method is described for reducing the number of degrees of freedom in a Markov source model for the purpose of image compression, comprising the steps of: (1) mapping the k-component Markov states into a smaller set of single-component states by a linear collapsing function, and (2) reducing each conditional probability distribution to a few parameters by matching them to known functions with the best fit. The Markov process is a basis of a general model with memory for compressing image sources. The image is viewed as a condition pixel source (X/S) producing a time sequence of N-bit pixels: XX{xn=ne{0,1,.....(2N-1)}} A kth-order Markov source model is
defined by the conditional probabilities {P(X/S)} and stationary probabilities {(S)}, where S is a Markov state consisting of k previous pels (X1,.X2,.Xj,.Xk), 1&j&k. Two Markov states, S1=(X1=a1,X2=a2,...,Xk=ak) and S2=(X1=b1,X2= b2,...,Xk=bk), are equivalent to each other if probability distribution P(X/S1) is the same as P(X/S2). Statistics for equivalent states can be gathered and combined as if they were virtually the same state. A k-component state S can be reduced to a single component N-bit equivalent state Sv where SvX{si:ie{0,1,...,(2N-1)}} by mapping S to Sv using a

"collapsing" function F: Sv=F(S)=F(X1,X2,...,Xk). If a linear function

Sv=A1X1+A2X2+...AkXk is chosen as F, states which are not entirely equivalent might be m...