Video Compression via Frequency Domain Block Matching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding techniques fail to achieve low complexity while delivering high-quality motion estimation in the frequency domain, leading to inefficiencies in data compression and transmission.

Innovation Solution

A method for performing video coding, including quantization and prediction, based on a perceptual model that accounts for human sensitivity to spatial and temporal frequencies, using block matching in the frequency domain to select reference blocks for prediction, and dynamically adjusting compression parameters such as distance and screen settings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If block matching is performed in the spatial domain using conventional techniques, then motion estimation can be achieved, but computational complexity increases and video quality deteriorates

Engineering Contradiction:
Improvecomputational complexityVSAvoidvideo quality
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The patent transforms the block matching operation from the spatial domain to the frequency domain by applying discrete cosine transforms (DCT) to both the reference and current blocks. This dimensional change allows motion estimation to be performed on frequency coefficients rather than spatial pixels, reducing computational complexity while maintaining or improving matching accuracy through better representation of perceptually important features.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent changes the representation parameters of the video blocks from spatial pixel values to frequency domain coefficients through DCT transformation. This parameter transformation enables the block matching algorithm to operate in a different mathematical space where computational operations are more efficient and produce better motion estimation results for the same computational cost.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If compression is applied to reduce bitrate, then data transmission efficiency improves, but video quality may deteriorate

Engineering Contradiction:
Improvedata transmission efficiencyVSAvoidvideo quality
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies quantization to the frequency domain coefficients, transforming them from precise floating-point values to integer values with reduced precision. This parameter change enables significant compression of the data while the human visual system's insensitivity to high-frequency variations means that video quality is preserved at lower bitrates.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies different quantization levels to different frequency coefficients, using coarser quantization for high-frequency components that are less perceptible to humans and finer quantization for low-frequency components that contain the most important visual information. This local differentiation of quality maintains perceived video quality while maximizing compression efficiency.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP2629523B1Data compression for video
Publication Date: 2021.04.07 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP2629523B1 patent drawingFigure 1
  • EP2629523B1 patent drawingFigure 1a
  • EP2629523B1 patent drawingFigure 1b

AI summary

The invention relates to a method of compressing a video signal for transmission to a decoder of a receiving terminal, the method comprising: receiving a video signal comprising a plurality of frames, each frame comprising a plurality of portions of video data; for each of a plurality of target ones of said portions, generating difference data representing a difference between the target image portion and a respective reference portion; and outputting the difference data in an encoded bitstream; wherein the method comprises, prior to generating the difference data, modifying at least one of said target and reference portions based on a measure of human sensitivity to the video data, so as to reduce a number of bits required to encode the difference data.