HDR Image Generation Using Exposure-Specific Neural Denoisers

Solution Overview

Problem

Generating high-dynamic range (HDR) images without noise or distortion in the extremes of the dynamic range is a difficult and time-consuming process, especially when capturing multiple images with different exposure levels, which can introduce motion blur and alignment issues.

Innovation Solution

The use of a system that combines multiple images or a single image with denoisers associated with different exposure levels, employing neural networks for denoising and exposure compensation to generate a weighted blend of pixel values, and optionally incorporating supplemental images for extreme highlight detail recovery.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Illumination intensity

If multiple images with different exposure levels are captured to generate HDR image, then dynamic range is improved, but time consumption and complexity increase

Engineering Contradiction:

Improvedynamic rangeVSAvoidtime consumption

Core Design Contradiction:

Illumination intensityVSLoss of time

Solution Approach 1:

The patent uses a single input image and generates multiple virtual images with different exposure levels by applying denoisers trained on synthetic data. Instead of capturing multiple real images with different exposures, the system creates synthetic versions of a single image that simulate various exposure conditions, thereby reducing time consumption while maintaining HDR quality.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces the mechanical process of capturing multiple physical images with different exposure settings using a computational approach. By using neural network denoisers trained on synthetic data, the system computationally generates multiple exposure versions from a single image, substituting physical capture operations with digital processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Illumination intensity

If multiple images with different exposure levels are captured to generate HDR image, then dynamic range is improved, but device complexity increases

Engineering Contradiction:

Improvedynamic rangeVSAvoidprocess complexity

Core Design Contradiction:

Illumination intensityVSDevice complexity

Solution Approach 1:

The patent uses a single input image and generates multiple virtual images with different exposure levels by applying denoisers trained on synthetic data. Instead of capturing multiple real images with different exposures, the system creates synthetic versions of a single image that simulate various exposure conditions, thereby reducing time consumption while maintaining HDR quality.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent changes the exposure parameter computationally by applying different denoisers that simulate various exposure levels. Rather than physically adjusting camera exposure settings, the system modifies the image data through neural network processing to create virtual images with different exposure characteristics, simplifying the overall process.

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If multiple images are captured to capture extreme details, then image quality is improved, but motion blur and alignment issues occur

Engineering Contradiction:

Improveimage qualityVSAvoidmotion blur

Core Design Contradiction:

Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The patent uses a single input image and generates multiple virtual images with different exposure levels by applying denoisers trained on synthetic data. Instead of capturing multiple real images with different exposures, the system creates synthetic versions of a single image that simulate various exposure conditions, thereby reducing time consumption while maintaining HDR quality.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces the mechanical process of capturing multiple physical images with different exposure settings using a computational approach. By using neural network denoisers trained on synthetic data, the system computationally generates multiple exposure versions from a single image, substituting physical capture operations with digital processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11776091B1Method and device for generating a high-dynamic range (HDR) image

Publication Date: 2023.10.03 APPLE INC

AI summary

In one implementation, a method includes obtaining an image. The method includes generating a plurality of denoised images by denoising the image using a respective plurality of denoisers, wherein the plurality of denoisers includes a first denoiser associated with a first level of exposure and a second denoiser associated with a second level of exposure. The method includes generating a combined image by combining the plurality of denoised images.