Digital Camera Image Deblurring via Sharpest Frame Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Consumer digital cameras and camera phones often produce image blur due to insufficient light or high zoom factors, and existing deblurring algorithms fail to accurately estimate the blur function, resulting in artifacts in deblurred images.
Innovation Solution
Select the sharpest image from a sequence captured around the blurred image and estimate the non-parametric blur function using a down-sampled version of the blurred image, employing an iterative Least Mean Squares (LMS) adaptive filtering algorithm to eliminate blur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If classical image deblurring algorithms are used, then image blur can be removed, but artifacts appear in the deblurred images due to inaccurate blur function estimation
Solution Approach 1:
The patent applies preliminary action by capturing a sequence of images before and after the blurred image to obtain reference images that are free from motion blur. These reference images are prepared in advance to enable accurate blur function estimation through correlation analysis, thereby avoiding the artifacts that plague conventional deblurring methods.
Solution Approach 2:
The patent replaces the mechanical/parametric approach of assuming a specific blur function form with a data-driven spectral correlation method. Instead of mechanically estimating blur parameters from the blurred image alone, the system uses spectral analysis of the relationship between reference and blurred images to automatically determine the blur function, substituting manual parameter estimation with an automated mathematical approach.
2Device complexity
If parametric blur function estimation is used, then the deblurring process can be simplified, but accuracy is lost when hand motion deviates from linear paths
Solution Approach 1:
The patent fundamentally changes the parameter estimation approach by moving from parametric models (which assume linear motion and require estimating only length and direction) to a spectral correlation method that operates in the frequency domain. This parameter transformation allows the system to capture complex non-linear hand motion patterns without increasing algorithmic complexity, as the spectral method naturally handles arbitrary motion trajectories through correlation analysis.
Data Source
AI summary
Deblurring of digital camera images by estimating the blur function from an image extracted from a video sequence taken about the time of an image capture. The extracted image is selected to be the sharpest of the images of the video sequence, and comparison of this sharpest image with the (down-sampled) captured image provides blur function iterative estimation.