Pixel-Wise Texture Filters for Reliable Image Super-Resolution

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image super-resolution methods, such as Super-Resolution Convolutional Neural Networks (SRCNN), Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN), and Very Deep networks for Super-Resolution (VDSR), struggle to reliably process images with varying texture features, particularly face images, leading to issues like fuzzy edges and high-frequency noise.

Innovation Solution

An image processing method that determines a unique filter for each pixel based on its texture features, using local texture images and weight values to perform super-resolution processing, thereby enhancing the reliability and efficiency of image enhancement.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a single super-resolution neural network is used for all image types, then the processing is simple and fast, but the processing reliability is poor for different texture types

Engineering Contradiction:
Improveimage processing reliabilityVSAvoidprocessing system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the image into multiple local image blocks and processes each block independently with tailored filter parameters. This segmentation allows different processing strategies for different texture regions, improving reliability without requiring a completely separate system for each image type.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by determining different filter parameters for different local image blocks based on their texture characteristics. Each block receives processing optimized for its specific texture type (e.g., face, building, natural scenery), which improves processing reliability while maintaining a unified processing framework.

Inventive Principle:
Principle #3Local quality

2Manufacturing precision

If iterative back-projection is used to recover high-frequency details, then high-frequency information is recovered, but fuzzy edges and high-frequency noise are introduced

Engineering Contradiction:
Improvehigh-frequency detail recoveryVSAvoidfuzzy edges and noise
Core Design Contradiction:
Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The patent determines different filter parameters for different local image blocks based on their texture characteristics. For blocks with edges or high-contrast features, the filter parameters are adjusted to preserve sharpness and avoid fuzzy edges. For flat-texture regions, the parameters are tuned to suppress noise while recovering high-frequency details, thus resolving the contradiction between detail recovery and noise introduction.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes filter parameters dynamically based on local image characteristics. By analyzing texture features of each local block and adjusting filter parameters accordingly, the system can recover high-frequency details in appropriate regions while avoiding the introduction of fuzzy edges and noise in regions where they would be harmful.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If texture-structure constraints are applied to recover high-frequency information, then high-frequency details are improved, but the processing complexity increases

Engineering Contradiction:
Improvehigh-frequency information recoveryVSAvoidprocessing algorithm complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent divides the image into multiple local image blocks and applies texture-structure constraints independently to each block. This segmentation reduces the overall computational complexity compared to applying constraints to the entire image at once, while still achieving effective high-frequency information recovery in each local region.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies texture-structure constraints selectively to local image blocks that benefit from high-frequency recovery, rather than uniformly to the entire image. This partial application reduces processing complexity while maintaining the benefits of high-frequency detail recovery where most needed.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3992903B1Image processing method, apparatus, and device
Publication Date: 2025.12.24 HUAWEI TECH CO LTD
  • EP3992903B1 patent drawingFigure 1
  • EP3992903B1 patent drawingFigure 2
  • EP3992903B1 patent drawingFigure 3

AI summary

This application provides an image processing method, apparatus, a device, and the like. In the method, a special image filter is generated, and super-resolution is performed on an image based on the image filter, thereby improving an image super-resolution effect. The image filter includes filter parameters corresponding to each pixel in an image that requires super-resolution processing, and pixels with different texture features correspond to different filter parameters. The image super-resolution method, apparatus, device, and the like may be applied to various scenarios such as a video, a game, and photographing, to improve an image effect in these scenarios, and enhance user experience.