Single-Image Deblurring via Horizontal-Vertical Blur Decomposition

Solution Overview

Problem

Existing image deblurring methods, particularly using U-Net architectures, struggle to effectively handle complex blurs such as motion blur and camera shake due to the entangled nature of blur characteristics, leading to reduced image quality and performance in computer vision tasks.

Innovation Solution

A single image deblurring method via vertical and horizontal decomposition using a one-encoder-two-decoder network system, where a convolution kernel is rotated and reused across decoders to separate and remove transverse and longitudinal blurs without explicit supervision, enhancing network efficiency and performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a single decoder architecture is used in U-Net, then the network structure is simple, but it cannot effectively handle complex blurs like object movement and camera shake

Engineering Contradiction:

Improvedeblurring performanceVSAvoidnetwork architecture complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The decoder is segmented into multiple specialized decoders (first decoder, second decoder, third decoder) that each handle specific blur directions or types. This segmentation allows the network to effectively process complex blurs by dividing the task into manageable directional components while maintaining overall architectural simplicity through parameter sharing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements parameter sharing across multiple decoders where the first decoder's parameters are shared by the second and third decoders. This universality allows a single set of learned parameters to serve multiple decoding functions for different blur directions, improving deblurring performance without proportionally increasing network complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If multiple independent decoders are used to handle different blur directions, then deblurring performance improves, but the number of parameters increases unnecessarily

Engineering Contradiction:

Improvedeblurring performanceVSAvoidnumber of parameters

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The patent implements parameter sharing across multiple decoders where the first decoder's parameters are shared by the second and third decoders. This universality allows a single set of learned parameters to serve multiple decoding functions for different blur directions, improving deblurring performance without proportionally increasing network complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

Instead of creating entirely independent decoders for different blur directions, the patent uses parameter sharing where the second and third decoders copy/reuse the parameter structure from the first decoder. This copying approach reduces the total number of parameters while still enabling specialized processing for different blur types.

Inventive Principle:

Principle #26Copying

3Ease of operation

If the network processes complex blurs without external supervision, then ease of operation improves, but difficulty in detecting and measuring blur components increases

Engineering Contradiction:

Improvetraining convenienceVSAvoidblur component separation

Core Design Contradiction:

Ease of operationVSDifficulty of detecting and measuring

Solution Approach 1:

The decoder is segmented into multiple specialized decoders (first decoder, second decoder, third decoder) that each handle specific blur directions or types. This segmentation allows the network to effectively process complex blurs by dividing the task into manageable directional components while maintaining overall architectural simplicity through parameter sharing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The network performs self-service by automatically decomposing and processing different blur components without requiring external supervision or manual intervention. The multiple decoders with parameter sharing enable the system to autonomously handle complex blurs, making the operation easier while effectively detecting and measuring blur components through the specialized decoder structures.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS12632940B2Single image deblurring method via horizontal and vertical decomposition and network system therefor

Publication Date: 2026.05.19 KOREA UNIV RES & BUSINESS FOUND

AI summary

A single image deblurring method via vertical and horizontal decomposition includes receiving, by an input unit of a deblurring network system, a single image; extracting, by an encoder, features from the received single image; decoding, by a horizontal decoder, the features provided from the encoder using a first convolution kernel to generate a first residual image; decoding, by a vertical decoder, the features provided from the encoder using a second convolution kernel generated by rotating the first convolution kernel to generate a second residual image; and synthesizing, by a synthesis unit, the first residual image and the second residual image into the received single image to generate a deblurred image.