Point Spread Function Estimation for Image Shake Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital photographing apparatuses face challenges in correcting image shake caused by user tremors, as existing methods are inefficient in accurately estimating and compensating for the point spread function (PSF) during image capture.
Innovation Solution
A method and apparatus that estimate the point spread function (PSF) by calculating global motion between short-exposure and long-exposure images, applying a difference-of-Gaussian filter, converting images into n-level images, correlating them to generate a correlation map, and deducing the PSF using a threshold value determined by a gradient descent method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional shake correction methods are used, then image shake can be corrected to some extent, but the estimation of point spread function is inaccurate and inefficient
Solution Approach 1:
The patent segments the image processing into multiple stages: global motion compensation first, then local motion estimation using PSF. The image is processed through multiple filtering stages (Gaussian filtering, band-pass filtering) and the PSF estimation is performed separately for different frequency components, allowing accurate yet efficient shake correction
Solution Approach 2:
The patent performs global motion compensation as a preliminary step before PSF estimation. By removing the global motion component first, the subsequent PSF estimation focuses only on local shake patterns, improving both accuracy and efficiency of the overall shake correction process
2Reliability
If band pass filter is applied to process image data, then noise and blur are reduced, but processing complexity increases
Solution Approach 1:
The patent combines multiple filtering operations (Gaussian filtering with different standard deviations, band-pass filtering) into a unified processing pipeline. The difference-of-Gaussian filters are merged to create the band-pass filter response, reducing overall processing complexity while maintaining image quality
Solution Approach 2:
The patent applies different filtering strengths to different frequency components of the image. The band-pass filter targets specific frequency ranges where shake patterns appear, applying stronger filtering only where needed rather than uniformly across the entire image, thus improving quality without excessive complexity
Data Source
AI summary
A method of estimating a point spread function (PSF) includes: estimating a global motion between a short-exposure image and a long-exposure image that are continuously captured using different exposure times, and compensating for the global motion; calculating a first resultant image by applying a band pass filter to the short-exposure image; calculating a second resultant image by applying the band pass filter to the long-exposure image; converting the first resultant image and the second resultant image into n-level images, where n is an odd natural number greater than or equal to 3, by deducing a first n-level resultant image and a second n-level resultant image from the first resultant image and the second resultant image, respectively; correlating the first n-level resultant image and the second n-level resultant image, and calculating a correlation map; and deducing the PSF from the correlation map.