Convolution Kernels for Motion Blur De-blurring

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current computational optics lack accurate mathematical models to capture motion without blur, especially for dynamic image formation, limiting the effectiveness of computational image processing in removing motion blur, as they can only simulate phenomena explained by geometric optics and not the wave nature of light.

Innovation Solution

The solution involves obtaining velocity vector data for moving subjects to compute convolution kernels, which are applied to image data to de-blur images by adjusting pixel values based on the motion trajectory and velocity profiles, effectively simulating the wave nature of light and improving image clarity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If computational optics methods are used to process light, then processing functionality can be performed computationally rather than mechanically, but accurate mathematical models to capture motion without blur are lacking

Engineering Contradiction:
Improvecomputational processing capabilityVSAvoidimage clarity
Core Design Contradiction:
Extent of automationVSManufacturing precision

Solution Approach 1:

The patent changes the parameter representation from simple geometric optics to a more comprehensive model that incorporates wave nature of light parameters. By using convolution kernels that account for both geometric and wave optical phenomena, the system achieves accurate motion blur correction while maintaining computational efficiency.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces convolution kernels as an intermediary mathematical model that bridges the gap between geometric optics computations and wave nature phenomena. These kernels act as a mediator that enables computational optics to simulate diffraction and interference effects without requiring complex wave equation solutions.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If current computational models based on geometric optics are used, then computations can be performed efficiently, but phenomena having to do with the wave nature of light like interference and diffraction cannot be simulated

Engineering Contradiction:
Improvecomputation speedVSAvoidsimulation capability
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal computational model that can handle both geometric optics and wave optics phenomena through the same convolution framework. The convolution kernels are designed to be multi-functional, capable of representing different optical effects (geometric projection, diffraction, interference) within a unified computational structure.

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

3Ease of operation

If motion blur correction is attempted without accurate velocity vector data, then processing can proceed with available image data, but the ability to reconstruct sharp images from blurred data is limited

Engineering Contradiction:
Improveprocessing simplicityVSAvoidimage reconstruction accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent performs preliminary estimation of velocity vectors using available image data before applying the convolution correction. By estimating motion parameters first and then using them to guide the deconvolution process, the system achieves accurate image reconstruction without requiring precise prior knowledge of motion trajectories.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2420970B1Motion blur modeling for image formation
Publication Date: 2020.03.18 HONEYWELL INTERNATIONAL INC
  • EP2420970B1 patent drawingFigure 1
  • EP2420970B1 patent drawingFigure 2
  • EP2420970B1 patent drawingFigure 3

AI summary

The present disclosure includes motion blur modeling methods and systems for image formation. One motion blur modeling method for image formation includes obtaining image data utilized to form a particular image taken by a camera of a subject, obtaining velocity vector data for the subject at the time the image was taken, defining a convolution kernel for use formation of the particular image and based upon the velocity vector data, and applying the convolution kernel to the image data to produce a de-blurred set of image data utilized to form the particular image.