Raw Image Reconstruction Using Frequency-Separated ISP Processing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image reconstruction methods struggle to reverse non-linear and spatially varying image signal processor (ISP) operations to accurately reconstruct raw image data, especially in consumer electronics, due to the intensive computing power required and the difficulty in inverting non-linear ISP processes.

Innovation Solution

A method and system for reconstructing raw image data by generating low-frequency and high-frequency images from an initial image, applying linear estimation and sparse interpolation techniques, using decomposition and reconstruction matrices to generate a reconstructed raw image from rendered images, with dynamic range compression and inverse processing to achieve accurate raw image reconstruction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If non-linear ISP operations are applied to render final images, then image quality and visual accuracy are improved, but the ability to reconstruct raw image data is degraded

Engineering Contradiction:

Improveraw image reconstruction accuracyVSAvoidISP processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by performing linear estimation and sparse interpolation on frequency-separated image components before final reconstruction. The image is first decomposed into low-frequency and high-frequency components, with the high-frequency components being linearly estimated from metadata, then sparsely interpolated to reconstruct the full high-frequency content. This preliminary processing enables accurate raw image reconstruction despite non-linear ISP operations.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent applies segmentation by dividing the image reconstruction process into distinct frequency components. The rendered image is separated into low-frequency and high-frequency parts, allowing different reconstruction strategies to be applied to each component. The high-frequency components are reconstructed using linear estimation from metadata, while low-frequency components are preserved from the original rendered image, enabling accurate overall reconstruction.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If full-resolution metadata is stored for accurate raw reconstruction, then reconstruction accuracy is improved, but storage requirements and computational overhead increase

Engineering Contradiction:

Improveraw reconstruction accuracyVSAvoidmetadata size

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies the taking out principle by extracting only the essential high-frequency information from the full-resolution image and storing it as compact metadata. Instead of storing complete high-frequency image data, the method extracts and stores only the necessary components that can be used to linearly estimate and reconstruct the full high-frequency content, significantly reducing metadata size while maintaining reconstruction accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by storing only a subset of the full image information—specifically, the high-frequency components—are stored as compact metadata, while the low-frequency components are preserved from the original rendered image. This partial storage approach achieves accurate reconstruction without the computational overhead of storing and processing complete high-resolution metadata.

Inventive Principle:

Principle #16Partial or excessive action

3Reliability

If non-linear color correction and tone mapping are applied, then visual image quality is improved, but reversibility for raw reconstruction is degraded

Engineering Contradiction:

Improveimage visual qualityVSAvoidreconstruction process simplicity

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The patent applies preliminary action by performing frequency decomposition and linear estimation before the final reconstruction step. The rendered image is first separated into frequency components, with high-frequency components being linearly estimated from compact metadata. This preliminary linear processing enables the system to bypass the complexity of inverting non-linear tone mapping and color correction operations, simplifying the overall reconstruction process while maintaining visual quality.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces frequency-domain decomposition as an intermediary step between the rendered image and the reconstructed raw image. By transforming the problem into the frequency domain and using linear estimation on high-frequency components, the method creates an intermediate representation that bridges the gap between non-linear rendered images and linear raw images, making the reconstruction process more straightforward.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250225610A1Raw image data reconstruction system and method

Publication Date: 2025.07.10 DOLBY LABORATORIES LICENSING CORP

AI summary

A method for reconstructing a raw image, including: generating a low-frequency image and a high-frequency image from an initial image; linearly estimating the high-frequency image to generate a reconstructed high-frequency image; sparsely interpolating the low-frequency image to generate a reconstructed low-frequency image; and generating a reconstructed raw image from the reconstructed low-frequency image and the reconstructed high-frequency image.