Deconvolution Kernel Gradient Operator Regularization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing deconvolution technologies, such as Wiener filtering, suffer from noise-induced ringing effects that degrade the quality of recovered sharp images due to their reliance on signal-to-noise ratio (SNR) as a regularization constraint.

Innovation Solution

A deconvolution kernel is determined based on a gradient operator and a convolution kernel, using the gradient operator as a regularization constraint to prevent noise interference and improve image recovery quality, specifically through the introduction of singular value decomposition to decompose the deconvolution kernel into one-dimensional kernels for efficient computation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If Wiener filtering with SNR-based regularization constraint is used for deconvolution, then the computation is simplified, but noise-induced ringing effects occur that degrade image recovery quality

Engineering Contradiction:
Improvecomputation simplicityVSAvoidimage recovery quality
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent changes the regularization constraint parameter from SNR (signal-to-noise ratio) to gradient operator. This parameter substitution fundamentally alters the deconvolution approach, replacing the noise-sensitive SNR constraint with a gradient-based constraint that promotes piecewise smooth solutions. The gradient operator constraint is formulated as minimizing the L2-norm of image gradients, which effectively suppresses noise-induced ringing while preserving edges, thus resolving the contradiction between computational simplicity and image recovery quality.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If gradient operator is introduced as regularization constraint instead of SNR, then noise interference is prevented and image recovery quality improves, but computational complexity increases

Engineering Contradiction:
Improveimage recovery qualityVSAvoidcomputation complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by decomposing the two-dimensional deconvolution problem into multiple one-dimensional subproblems. The gradient operator constraint is separated into horizontal and vertical gradient components, allowing the optimization to be performed independently in each dimension. This segmentation reduces the overall computational complexity while maintaining the quality improvements achieved through gradient-based regularization.

Inventive Principle:
Principle #1Segmentation

3Productivity

If deconvolution kernel is decomposed into one-dimensional kernels, then computational efficiency is improved, but the accuracy of image recovery may be affected

Engineering Contradiction:
Improvecomputational efficiencyVSAvoidimage recovery accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent segments the deconvolution kernel into separable one-dimensional kernels along horizontal and vertical directions. This segmentation allows the computationally intensive two-dimensional convolution operation to be replaced by sequential one-dimensional convolutions, significantly improving efficiency. The gradient operator constraint ensures that this segmentation does not compromise recovery accuracy by maintaining the piecewise smooth property in both dimensions independently.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent merges the deconvolution operation with the gradient-based regularization constraint into a unified optimization framework. By combining these two functions (deconvolution and regularization) into a single energy minimization problem, the patent ensures that the separable kernel approximation maintains accuracy while achieving computational efficiency through the coordinated action of both components.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP2993642B1Method and apparatus for generating sharp image based on blurry image
Publication Date: 2017.10.11 HUAWEI TECH CO LTD
  • EP2993642B1 patent drawingFigure 1
  • EP2993642B1 patent drawingFigure 2(a)~2(c)
  • EP2993642B1 patent drawingFigure 3~4

AI summary

The present invention discloses a method for generating a sharp image based on a blurry image. The method includes: acquiring pixel values of pixels in the blurry image, and a convolution kernel of the blurry image; determining a deconvolution kernel of the blurry image based on a preset image gradient operator and the convolution kernel; determining pixel values of pixels in the sharp image based on the deconvolution kernel and the pixel values of the pixels in the blurry image; and generating the sharp image based on the pixel values of the pixels in the sharp image. A deconvolution kernel in the embodiments of the present invention is determined based on a gradient operator and a convolution kernel; in other words, the deconvolution kernel introduces the gradient operator as a regularization constraint, which prevents noise from affecting an image recovery process, and improves the quality of a recovered sharp image.