Authors:

ILo Chen

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

It is easy to get bad-quality pictures in everyday life. For example, nonsufficient time of exposure or shaking hands will cause the image to be noisy or blurry. As a result, there have been many different image restoration algorithms using computer vision and have been broadly applied in the camera industries. One well-known study in this area is the Plug-and-Play Alternating Direction Method of Multiplier framework (PnP ADMM) that can restore an imperfect image regardless of the restoration tasks. This PnP ADMM is an iterative method based on an objective function comprising of two terms, image reformation model and natural image prior, that need to be determined. Image reformation model is determined based on the restoration task, and natural image prior can be determined based on personal preferences. There have been multiple natural image priors proposed from hand-crafted to learning-based, but limited studies were done on the pros and cons of them, which makes choosing which prior to use a difficult decision. We compare the computational cost, convergence, accuracy, and robustness for image deblurring task between different natural images priors in our research and provide guidelines on how to select the most efficient natural image prior to PnP ADMM for future applications.

Source:

Purdue University / 2023

Topics:

No topics listed

Co-authors:

ILo Chen