Pixel Quantization Rounding Mode Selection for Temporal Coherence

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for quantizing floating values of pixels in images often result in temporal quantization incoherencies, particularly when converting dynamic ranges, leading to inefficiencies in coding high dynamic range images and increased residual prediction errors.

Innovation Solution

A method that involves obtaining a test image by reducing the dynamic range of a second image, quantizing its floating point values, and then using these quantized values to determine the rounding of floating point values in the original image, selecting between rounding to the greater, lesser, or nearest whole number based on differences and thresholds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional rounding methods are used to quantize floating values of pixels, then the quantization process is simple and fast, but temporal quantization incoherencies appear between successive images

Engineering Contradiction:
Improvequantization speedVSAvoidtemporal coherence
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies preliminary action by pre-determining the rounding mode for each pixel position based on a test image before actually quantizing the target image. The rounding mode selection is performed in advance by analyzing a test image with the same dynamic range reduction, and then this pre-selected rounding mode is applied to the target image, ensuring temporal coherence without sacrificing quantization speed.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses copying by creating a test image that replicates the processing conditions (dynamic range reduction) applied to the target image. This test image serves as a model to determine the appropriate rounding modes, allowing the system to learn from the test case and apply the same rounding behavior to the target image, thereby ensuring temporal consistency across successive images.

Inventive Principle:
Principle #26Copying

2Adaptability or versatility

If dynamic range reduction is applied to convert high dynamic range images to standard dynamic range, then the images can be displayed on standard screens, but temporal quantization incoherencies and increased residual prediction errors occur

Engineering Contradiction:
Improvedisplay compatibilityVSAvoidcoding efficiency
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent applies local quality by selecting different rounding modes (rounding to greater whole number, lesser whole number, or nearest whole number) for different pixel positions based on local characteristics observed in the test image. This localized rounding mode selection optimizes the quantization quality for each region while maintaining temporal coherence, thereby reducing residual prediction errors and preserving coding efficiency.

Inventive Principle:
Principle #3Local quality

3Measurement precision

If floating values are used to represent pixel components with greater than 8 bits dynamic range, then the dynamic range representation capability is improved, but quantization complexity increases

Engineering Contradiction:
Improvedynamic range representationVSAvoidquantization process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent reduces quantization complexity by performing the complex rounding mode selection in advance on a test image. The preliminary analysis determines the appropriate rounding mode for each pixel position, and this pre-computed information is then reused for the actual target image quantization, avoiding repeated complex decisions and simplifying the real-time processing.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3074946B1Method and device for quantising the floating value of a pixel in an image
Publication Date: 2018.10.17 THOMSON LICENSING SA
  • EP3074946B1 patent drawingFigure 1~4
  • EP3074946B1 patent drawingFigure 5~6
  • EP3074946B1 patent drawingFigure 7~8

AI summary

The invention relates to a method and device for quantising the floating value of a pixel of an image by rounding either to a lesser whole number, to a greater whole number, or to the whole number closest to this floating value. The method is characterised in that the selection of rounding this floating value is determined based on a test value.