Systems and methods for distributed quantization of multimodal images

Patent metadata

Field	Value
Publication number	US-11051030-B2
Application number	US-202016734885-A
Country	US
Kind code	B2
Filing date	Jan 6, 2020
Priority date	Sep 30, 2016
Publication date	Jun 29, 2021
Grant date	Jun 29, 2021

Abstract

A system and method for implementing a distributed source coding quantization scheme is provided. In one example, two independent but statistically correlated data sources can be asymmetrically compressed so that one source is compressed at a higher ratio than the other. The resulting signals are transmitted and decoded by a receiver. The highly compressed source can utilize the non-highly compressed source as side information so as to minimize the compression loss associated with the higher compression ratio. A conditional codebook can be created that not only depends on the highly compressed quantizer, but also depends on the quantized symbol received from the non-highly compressed data source.

First claim

The invention claimed is:

1. A method for decoding quantized signals, the method comprising:
at a first decoder, wherein the first decoder comprises a memory that stores a plurality of codebooks generated during a training process, the training process comprising:
receiving a first training sequence, wherein the first training sequence is based on a first data source;
generating a first codebook based on the received first training sequence;
receiving a second training sequence, wherein the second training sequence is based on a second data source;
determining an initial codebook based on the second training sequence;
determining a first optimized quantizer based on the initial codebook;
determining a first optimized codebook based on the determined first optimized quantizer;
determining a second optimized quantizer based on the determined first optimized codebook;
determining a second optimized codebook based on the determined second optimized quantizer; and
generating a second codebook based on the second optimized codebook.

2. The method of claim 1 , comprising:
at a second decoder:
receiving a first quantized collection of bits, wherein the first quantized collection of bits is based on data acquired by a first data source, and wherein the first quantized collection of bits represents a compressed first file;
selecting a first codeword from a first codebook based on the first received quantized collection of bits; and
reconstructing the data acquired by the first data source using the selected first codeword and the received first quantized collection of bits;

and wherein at the first decoder:
receiving a second quantized collection of bits, wherein the second quantized collection of bits is based on data acquired by a second data source, wherein the first data source and second data source acquire data independently from one another, wherein the second quantized collection of bits represents a compressed second file, wherein an amount of compression of the compressed second file is greater than an amount of compression of the compressed first file, and wherein the data acquired by the first data source and the data acquired by the second data source are statistically correlated with respect to one another;
selecting a codeword from the second codebook based on the second received quantized collection of bits and the first received quantized collection of bits, wherein the second codebook is based on the first codebook and a relationship between the first quantized collection of bits and the second quantized collection of bits; and
reconstructing the data acquired by the second data source based on the received second quantized collection of bits and the selected second codeword.

3. The method of claim 1 , wherein determining the initial codebook includes applying a Lloyd algorithm to the second training sequence.

4. The method of claim 1 , wherein generating the first codebook includes applying a Lloyd algorithm to the first training sequence.

5. The method of claim 1 , wherein the training process further comprises:
converting the first training sequence into a first set of one or more image patches; and
converting the second training sequence into a second set of or more image patches.

6. The method of claim 2 wherein the data acquired by the first source is transmitted at a higher rate than the data acquired by the second source.

7. The method of claim 6 , wherein a number of bits of the first quantized collection of bits is greater than a number of bits of the second quantized collection of bits.

Assignees

• Mitre Corp

Inventors

• Taylor Jr Robert M
• Woodard Jeffrey P

Classifications

• H04N19/132

  Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking · CPC title

• H04N19/94

  Vector quantisation · CPC title

• H04N19/157

  Assigned coding mode, i.e. the coding mode being predefined or preselected to be further used for selection of another element or parameter · CPC title

• H04N19/179

  the unit being a scene or a shot · CPC title

• H04N19/105

  Selection of the reference unit for prediction within a chosen coding or prediction mode, e.g. adaptive choice of position and number of pixels used for prediction · CPC title