Gaussian Blur Approximation via Down-Sampling and Up-Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for implementing Gaussian blurs in graphics processing systems are computationally intensive and inefficient on current GPU architectures, which limits their ability to efficiently generate high-quality blur effects.
Innovation Solution
The method involves down-sampling an image using a truncated Gaussian filter and then up-sampling it back to the original size, utilizing a distortionless filter or linear interpolation, to approximate a Gaussian blur, reducing computational costs and optimizing performance on GPUs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional Gaussian blur implementation is used, then high-quality blur effect is achieved, but computational cost becomes excessively high
Solution Approach 1:
The patent segments the Gaussian blur operation into multiple passes of box blur operations. Instead of applying one large computationally intensive Gaussian filter, the method divides the blur into several smaller passes, each using a simple box filter that can be efficiently implemented on GPU hardware. This segmentation reduces the computational complexity from O(w) to O(1) per pass while maintaining visual quality through multiple iterations.
Solution Approach 2:
The patent replaces the expensive Gaussian filter with inexpensive box blur operations that can be rapidly executed. The box blur uses a simple averaging algorithm that is computationally cheap and can be disposed of (re-executed) multiple times to achieve the desired blur effect, rather than using a single expensive Gaussian convolution.
2Productivity
If repeated box blur operations are used to avoid computational cost, then processing speed improves, but current GPU architectures cannot efficiently implement the required row and column summing operations
Solution Approach 1:
The patent transforms the 2D box blur operation into a series of 1D operations that can be efficiently parallelized on GPU. By separating horizontal and vertical blur passes and using the GPU's natural parallel processing capability across pixels, the method avoids the need for complex sequential row and column summing operations that current GPU architectures struggle with.
3Use of energy by moving object
If a small blur radius is used, then computational cost decreases, but blur quality is insufficient
Solution Approach 1:
The patent applies multiple continuous passes of box blur operations rather than a single pass. Each pass contributes to the cumulative blur effect, and by chaining multiple passes together, the method achieves high-quality Gaussian-like blur without requiring a large blur radius in any single pass, thus maintaining low computational cost per pass while improving overall blur quality through the continuous application of the filter.
Data Source
AI summary
A Gaussian blur approximation is applied to an image by repeated down-sampling operations followed by an up-sample operation. By using a truncated Gaussian filter as the down-sample filter, the frequency spectrum removed during down-sampling operations closely approximates the frequency spectrum lost during a true Gaussian blur operation. While any “good” up-sample filter may be used, up-sampling via linear interpolation may be especially beneficial in systems having a dedicated graphics processing unit. One benefit of the described technique is that it is computationally less costly to implement than a Gaussian blur. Another benefit is that this computational benefit increases as the size of the blur increases—becoming significant even for small blurs. Yet another benefit of the invention is that a judicious selection of pixel address to convolve with the filter leads to substantially reduced number of texture lookups required to effect a convolution.