Purdue University Graduate School
Browse
- No file added yet -

Reducing Image Artifacts in Motion Blur Prevention

Download (32.19 MB)
thesis
posted on 2023-04-27, 17:27 authored by Zixun YuZixun Yu

Motion blur is a form of image quality degradation, showing as content in the image smearing and not looking sharp. It is usually seen in photography due to relative motion between the camera and the scene (either camera moves or objects in the scene move). It is also seen in human vision systems, primarily on digital displays.


It is often desired to remove motion blurriness from images. Numerous works have been put into reducing motion blur after the image has been formed, e.g., for camera-captured ones. Unlike post-processing methods, we take the approach to prevent/minimize motion blur for both human and camera observation by pre-processing the source image. The pre-processed images are supposed to look sharp upon blurring. Note that, only pre-processing methods can deal with human-observed blurriness since the imagery can't be modified after it is formed on the retina.


Pre-processing methods face more fundamental challenges than post-processing ones. A problem inherent to such methods is the appearance of ringing artifacts which are intensity oscillations reducing the quality of the observed image. We found that these ringing artifacts have a fundamental cause rooted in the blur kernel. The blur kernel usually have very low amplitudes in some frequencies, significantly attenuating the signal intensity in these frequencies when it convolves an image. Pre-processing methods can usually reconstruct the targeted image to the observer but inevitably lose energy in those frequencies, appearing as artifacts. To address the artifact issue, we present a few approaches: (a) aligning the image content and the kernel in the frequency domain, and (b) redistributing their intensity variations elsewhere in the image. We demonstrate the effectiveness of our method in a working prototype, in simulation, and with a user study.

Funding

Elements: Data: U-Cube: A Cyberinfrastructure for Unified and Ubiquitous Urban Canopy Parameterization

Directorate for Computer & Information Science & Engineering

Find out more...

III: Medium: Collaborative Research: Deep Generative Modeling for Urban and Archaeological Recovery

Directorate for Computer & Information Science & Engineering

Find out more...

EagleVision: Multi-view Preemptive Image-based Vision Correction

History

Degree Type

  • Doctor of Philosophy

Department

  • Computer Science

Campus location

  • West Lafayette

Advisor/Supervisor/Committee Chair

Daniel G Aliaga

Additional Committee Member 2

Bedřich Beneš

Additional Committee Member 3

Voicu S Popescu

Additional Committee Member 4

Manuel M Oliveira

Usage metrics

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC