Content-Adaptive Perceptual Quantization for HDR Image Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current technologies face challenges in efficiently processing high-dynamic-range images, particularly in terms of perceptual quantization, which affects image quality and compatibility across different display devices with varying dynamic ranges.

Innovation Solution

The implementation of Content-Adaptive Perceptually Quantized (CAQ) image processing techniques, including forward and backward reshaping functions, dynamic range non-linear scaling, and noise masking, to optimize bit depth allocation and image merging/blending, ensuring efficient encoding and decoding across a wide range of dynamic ranges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If perceptual quantization is applied to HDR images, then image quality is improved, but compatibility across different display devices deteriorates

Engineering Contradiction:
Improveimage qualityVSAvoiddisplay device compatibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the dynamic range into multiple bands (e.g., 5 bands) and applies different quantization strategies to each band. This allows fine-grained control over quantization precision in different luminance regions, improving overall image quality while maintaining adaptability to various display devices through selective application of quantization parameters.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs dynamic range non-linear scaling that adapts to different display device characteristics. By using content-adaptive techniques that adjust quantization parameters based on the actual content and target display properties, the system maintains high image quality across diverse display devices with varying dynamic ranges.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If high-precision floating-point formats are used for HDR images, then dynamic range representation is improved, but data transmission and processing efficiency deteriorates

Engineering Contradiction:
Improvedynamic range representationVSAvoiddata transmission efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent transforms HDR image data from high-precision floating-point format to optimized integer representations using perceptual quantization. By changing the numerical representation parameters based on human visual system characteristics and display device capabilities, the system achieves efficient data transmission and processing while preserving perceptual image quality.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent extracts and removes redundant high-precision information that is not perceptually significant. Through perceptual quantization, the system identifies and eliminates excessive precision in regions where the human visual system cannot distinguish differences, thereby reducing data size and improving transmission efficiency without noticeable quality loss.

Inventive Principle:
Principle #2Taking out (Extraction)

3Device complexity

If standard quantization methods are used for HDR images, then processing simplicity is maintained, but noise artifacts increase

Engineering Contradiction:
Improveprocessing simplicityVSAvoidnoise artifacts
Core Design Contradiction:
Device complexityVSObject-generated harmful factors

Solution Approach 1:

The patent applies local quality enhancement through content-adaptive quantization that adjusts parameters based on local image characteristics such as luminance level, noise content, and texture complexity. This allows the system to maintain processing simplicity while reducing noise artifacts by applying different quantization strategies to different regions of the image.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent converts potential noise artifacts into beneficial effects by using perceptual quantization that aligns with human visual system characteristics. By quantizing in a way that matches perceptual thresholds, the system transforms what would be noticeable noise into imperceptible quantization steps, effectively converting a harmful artifact into an imperceptible feature.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentEP3176749B1Efficient image processing on content-adaptive PQ signal domain
Publication Date: 2020.11.04 DOLBY LABORATORIES LICENSING CORP
  • EP3176749B1 patent drawingFigure 1A~1B
  • EP3176749B1 patent drawingFigure 2
  • EP3176749B1 patent drawingFigure 3A

AI summary

An image processing device receives one or more forward reshaped images that are generated by an image forward reshaping device from one or more wide dynamic range images based on a forward reshaping function. The forward reshaping function relates to a backward reshaping function. The image processing device performs one or more image transform operations on the one or more forward reshaped images to generate one or more processed forward reshaped images without performing backward reshaping operations on the one or more reshaped images or the one or more processed forward reshaped images based on the backward reshaping function. The one or more processed forward reshaped images are sent to a second image processing device.