Neural HDR Rendering With Pyramid Layers for Local Contrast

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing display technologies struggle to accurately render high dynamic range (HDR) content on displays with varying characteristics, leading to loss of detail and contrast, particularly in local image areas.

Innovation Solution

A neural network-based precision rendering process that splits an input image into a base layer and detail layer, using pyramid down-sampling and up-sampling techniques to preserve global and local contrast, employing neural networks for tone-mapping and edge-aware filtering to adapt the image to target display characteristics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional display management techniques are used to map HDR content to displays with varying characteristics, then the display process is simple and fast, but the rendered image loses detail and contrast, particularly in local image areas

Engineering Contradiction:
Improveimage rendering precisionVSAvoidprocessing complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the image processing into distinct functional components: a neural network module for tone mapping and a separate precision rendering module for detail preservation. This segmentation allows each module to specialize in specific tasks, improving overall rendering precision while managing complexity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension of processing by implementing multi-scale analysis through pyramid decomposition. The precision rendering process operates at multiple resolution levels, analyzing and reconstructing image details across different scales. This dimensional approach preserves local contrast and fine details that would be lost in single-scale traditional methods.

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

2Manufacturing precision

If neural network-based precision rendering is implemented to preserve local contrast and detail, then image quality is improved, but processing time and computational complexity increase

Engineering Contradiction:
Improvelocal contrast preservationVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-training the neural network with large datasets of HDR and SDR image pairs. The network learns optimal tone mapping transformations during an offline training phase, storing learned parameters and weights. During actual display management, the pre-trained network performs rapid inference, significantly reducing real-time processing time while maintaining high local contrast preservation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements partial action by applying the computationally intensive neural network processing selectively rather than to all image data uniformly. The precision rendering process focuses computational resources on regions requiring detail preservation while using more efficient methods for other areas, optimizing the balance between image quality and processing time.

Inventive Principle:
Principle #16Partial or excessive action

3Manufacturing precision

If pyramid down-sampling and up-sampling techniques are used to create base and detail layers, then global and local contrast are preserved, but the processing complexity and computational load increase

Engineering Contradiction:
Improvedetail preservationVSAvoidprocessing complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by separating the image into distinct base layer and detail layer components through pyramid decomposition. Each layer is processed independently with appropriate algorithms: the base layer handles global tone mapping while the detail layer preserves local features. This segmentation simplifies the overall processing complexity by breaking down the complex task into manageable, specialized sub-tasks.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies merging by combining the processed base layer and detail layer back together to form the final output image. The precision rendering module merges these layers in a way that preserves both global contrast from the base layer and local detail from the detail layer, achieving high-fidelity rendering while managing processing complexity through efficient combination operations.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4392928B1Neural networks for precision rendering in display management
Publication Date: 2025.11.12 DOLBY LABORATORIES LICENSING CORP
  • EP4392928B1 patent drawingFigure 1
  • EP4392928B1 patent drawingFigure 2~3
  • EP4392928B1 patent drawingFigure 4

AI summary

Methods and systems for precision rendering in display mapping using neural networks are described. Given an intensity input image, a sequence of neural networks comprising a pyramid-halving sub-network, a pyramid down-sampling sub-network, a pyramid-up-sampling sub-network, and a final-layer generation sub-network generate a base layer image and a detail layer image to be used in display mapping.