Edge-Preserving Image Smoothing via Resolution Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Retinex theory-based image processing methods face challenges in generating an illumination component image, particularly near edges, which can result in halo artifacts due to inadequate illumination component separation, requiring significant computation for edge preservation and smoothing.
Innovation Solution
An image processing device and method that includes a reducing unit, a reduced image smoothing unit, and an edge-preserving enlarging unit, which performs weighted addition of pixel values from reference pixels to determine target pixel values, setting weighting coefficients based on differences and distances to preserve edges during smoothing and enlargement, reducing computation and preventing halo artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a nonlinear filter (e.g. bilateral filter) is used for performing smoothing in areas other than edges while preserving the edges, then edge preservation is improved, but computation time increases hugely
Solution Approach 1:
The patent divides the image processing into two stages: first reducing the image to a smaller size where smoothing is computationally cheaper, then enlarging the smoothed image back to original size. This segmentation of the processing workflow allows edge preservation without the full computational cost of applying bilateral filtering at full resolution.
Solution Approach 2:
The patent performs the smoothing operation on a reduced-size image before enlargement, rather than on the full-resolution image. This preliminary action at lower resolution significantly reduces computation time while still achieving the desired edge preservation effect when the image is subsequently enlarged.
2Manufacturing precision
If the reference pixel range of the filter is widened to preserve edges, then edge preservation is improved, but the amount of computation increases hugely
Solution Approach 1:
The patent segments the processing by applying the filter with a wide reference pixel range on a reduced-size image, thereby reducing the total number of pixels that need to be processed. This maintains the benefit of wide-range filtering for edge preservation while reducing computational complexity through the size reduction.
3Productivity
If fast bilateral filtering is used to reduce computation time, then processing speed is improved, but circuit size increases
Solution Approach 1:
The patent performs smoothing on a reduced-size image before enlargement, which reduces the number of filtering operations needed. This preliminary action at lower resolution achieves the same visual effect with fewer computational resources, thereby reducing circuit size requirements while maintaining processing speed.
Data Source
AI summary
An image processing device for performing edge-preserving smoothing includes a reducing unit for reducing an input image; a reduced image smoothing unit for performing a smoothing processing on a reduced image generated by the reducing unit with edges preserved; and an edge-preserving enlarging unit for enlarging an image generated by the reduced image smoothing unit with the edges preserved. The edge-preserving enlarging unit performs a filtering processing by determining a pixel value of a target pixel for the filtering processing based on weighted addition of pixel values of respective reference pixels. Weighting coefficients for the pixel values of the respective reference pixels is set based on differences between a pixel value of the input image corresponding to the target pixel and pixel values of the respective reference pixels and based on distances between the target pixel and the respective reference pixels.