Sparse Convolution Kernel for Real-Time VR Blur Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of existing convolution methods for generating blurred images in real-time video applications, particularly in virtual reality systems, due to high computational requirements and the need for accurate point spreading function estimation, makes it challenging to achieve realistic blurring effects at high image refreshing rates.
Innovation Solution
An image processing method that uses a simplified model to determine a spreading parameter for each pixel based on the viewer's eye distance, dioptric power of a correction lens, and maximum accommodation power, reducing the number of pixels involved in the convolution operation, resulting in a sparse convolution kernel with fewer non-zero values, which allows for efficient blurring simulation with reasonable computation power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a convolution method with accurate point spreading function estimation is used to generate blurred images, then the blur rendering quality is improved, but the computational complexity increases significantly
Solution Approach 1:
The patent changes the parameters of the convolution operation by using a simplified PSF model that depends on fewer parameters (distance parameter z, dioptric power Pcorr, maximum accommodation power Pacc) instead of the full complex eye model. This allows maintaining acceptable blur quality while reducing the number of computational operations required.
Solution Approach 2:
The patent applies partial action by using a simplified PSF model that does not fully capture all optical effects (such as higher-order aberrations, pupil function variations, and complex eye model interactions). This partial modeling approach reduces computational requirements while providing sufficient blur rendering for the application context.
2Measurement precision
If a sophisticated eye model with large number of rays is used to estimate the point spreading function, then the measurement precision is improved, but the computational time increases
Solution Approach 1:
The patent creates a simplified copy of the PSF calculation that uses the same functional form but with reduced complexity parameters. Instead of tracing many rays through a complex eye model, the patent uses a direct calculation based on distance and dioptric power parameters, effectively copying the essential behavior with much less computation.
Solution Approach 2:
The patent extracts only the essential parameters needed for PSF estimation (distance parameter z and dioptric power Pcorr) from the complex eye model, discarding unnecessary computational steps. This extraction allows maintaining the core functionality of PSF calculation while eliminating redundant computations.
3Productivity
If the number of pixels involved in the convolution operation is reduced to decrease computational complexity, then the processing speed is improved, but the blur quality may deteriorate
Solution Approach 1:
The patent applies dynamics by making the PSF model adaptive to the distance parameter z and dioptric power Pcorr. The convolution kernel dynamically adjusts its characteristics based on the focal plane and optical conditions, allowing efficient computation at each pixel while maintaining quality through localized optimization rather than uniform processing.
Data Source
AI summary
In an image processing method, a first two-dimensional image (I(x, y)) of a three-dimensional scene is obtained. Each pixel location of a first image is associated with a respective distance parameter (z(x, y)) representing a distance of a viewer's eye to an object of the scene appearing at this pixel location in the first image. A respective spreading parameter (Φd(x, y)) is determined for each pixel location as a function of the distance parameter associated with this pixel location, of a dioptric power (Pcorr) of a portion of a correction lens simulated as placed along a ray extending between the viewer's eye and the object of the scene appearing at said pixel location in the first image, and of a simulated maximum accommodation power (Pacc) of the viewer's eye. A second, blurred two-dimensional image (I′(x, y)) is generated from the first image using a local convolution kernel varying across the image. This kernel can be made sparse, i.e. with a number of non-zero points much smaller than the number of points of a blur region whose dimension is given by the spreading parameter.