Accelerated predictive-transform
AssigneeResearch Foundation of the City University of New York
InventorsErlan H. Feria, Sos Agaian
Life of this patent
Estimating the term…
Abstract
The present invention is drawn to a method for digital image modeling in which the digital image is divided into digital image blocks. First, a coder input vector of the digital image blocks is received. Second, a coefficient vector is approximated, each coefficient or coefficient vector corresponding to one of the digital image blocks. The approximating step further comprises generating a transformation matrix that yields a coefficient vector from a product of the coder input vector and decomposing the transformation matrix into a product of multi-diagonal or sparse unitary transformation matrixes and an energy sequencer unitary transformation matrix.
Claims14
A method of digital image modeling, a digital image divided into a plurality of digital image blocks, comprising the steps of:
- receiving a coder input vector of a plurality of digital image blocks;
- approximating a coefficient vector, each coefficient of the coefficient vector corresponding to one of the digital image blocks;
- wherein the step of approximating the coefficient vector comprises:
- generating a forward transformation matrix that yields a coefficient vector from a product of the coder input vector;
- and
- decomposing the forward transformation matrix into a product of at least one of a multi-diagonal transformation matrix and a sparse unitary transformation matrix and an energy sequencer unitary transformation matrix.
A method of digital image modeling, a digital image divided into a plurality of digital image blocks, comprising the steps of:
- receiving a coefficient vector, each coefficient or coefficient vector corresponding to one of a plurality of digital image blocks;
- approximating a coder input vector of the digital image blocks;
- wherein the step of approximating the coder input vector comprises:
- generating a backward transformation matrix that yields a coder input vector from a product of the coefficient vector;
- and
- decomposing the backward transformation matrix into a product of at least one of a multi-diagonal transformation matrix and a sparse unitary transformation matrix and an energy sequencer unitary transformation matrix.
A method of digital image modeling, a digital image divided into a plurality of digital image blocks, comprising the steps of:
- receiving a predictive vector representing a plurality of digital image blocks immediately adjacent a coder input vector comprising a plurality of digital image blocks of a divided digital image;
- approximating a coefficient vector, each coefficient of the coefficient vector corresponding to one of the digital image blocks of the divided digital image;
- wherein the step of approximating the coefficient vector comprises:
- generating a forward prediction matrix that yields a coefficient vector from a product of the coder input vector;
- and
- decomposing the forward prediction matrix into a product of at least one of a multi-diagonal and a sparse unitary transformation matrix and an energy sequencer unitary transformation matrix and a cross variance matrix divided by the coder input vector's average power.
Dependencies are inferred from the claim text.