Superresolution Neural Network Blur Simulation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current super-resolution methods for image processing, such as bicubic interpolation, produce blurry and low-perceptual-quality results when attempting high digital zoom, and using depth maps alone can lead to unrealistic results with artifacts.

Innovation Solution

A method involving a superresolution neural network (SRNN) that incorporates focusing information and depth maps to generate super-resolved images, where focusing information is used to accurately simulate blur based on the focus distance, reducing artifacts and maintaining natural image effects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If classical interpolation methods like bicubic interpolation are used for digital zoom, then the processing is simple and fast, but the image quality becomes blurry and low perceptual quality

Engineering Contradiction:
Improveprocessing speedVSAvoidimage quality
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent replaces classical mechanical interpolation methods (bicubic interpolation) with a neural network-based super-resolution system. The neural network learns complex non-linear mappings from low-resolution to high-resolution images, capturing natural image structures and textures that linear interpolation cannot achieve, thereby significantly improving image quality while maintaining processing efficiency through optimized network architecture.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If depth map is used as constraint for super resolution, then the focus distance information is provided, but the results contain artifacts and appear unrealistic

Engineering Contradiction:
Improvefocus distance informationVSAvoidimage realism
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent merges multiple input sources including depth map, focusing information, and the original low-resolution image into a unified neural network processing pipeline. The neural network integrates these heterogeneous data types through learned feature representations, allowing the focusing information to guide the super-resolution process while the network's learned priors ensure natural-looking results free of artifacts, achieving both accurate focus modeling and visual realism.

Inventive Principle:
Principle #5Merging (Combining)

3Manufacturing precision

If focusing information is added to super resolution process, then the blur accuracy is improved and artifacts are reduced, but the device complexity increases

Engineering Contradiction:
Improveblur accuracyVSAvoidsystem complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent designs a neural network architecture that performs multiple functions within a single unified model: it processes the low-resolution image, incorporates focusing information, generates depth-related blur patterns, and produces the final high-resolution output. This multi-functional design eliminates the need for separate processing stages for focus modeling and super-resolution, reducing overall system complexity while maintaining high blur accuracy through the network's learned representations.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4398185A1Method for image processing
Publication Date: 2024.07.10 BEIJING XIAOMI MOBILE SOFTWARE CO LTD
  • EP4398185A1 patent drawingFigure 1~2
  • EP4398185A1 patent drawingFigure 3
  • EP4398185A1 patent drawingFigure 4

AI summary

Method for image processing, preferably implemented in a camera device, including capturing an image; acquiring focusing information indicating the focus distance of the image; providing the image and the focusing information to a superresolution neural network, and generating a super resolved image.