Polygon Kernels for Efficient Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image filtering techniques, such as those used for simulating camera defocus and depth of field, can be complex and computationally inefficient, especially when dealing with non-rectangular polygon kernels, as they require repetitive summing of pixel values, leading to excessive computational resource consumption.
Innovation Solution
The use of polygon transfer functions defined by cones, which allow for the creation of processing tables independent of the scale of the polygon, enabling efficient image processing operations that are independent of the kernel's size, using techniques like Brion's theorem to sum tangent cone transfer functions into polynomial expressions for convolution with arbitrary convex polygons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional filtering techniques are used for simulating camera defocus and depth of field, then image processing effects can be achieved, but computational complexity and resource consumption increase significantly
Solution Approach 1:
The patent segments the polygon kernel into multiple triangular facets that can be processed independently. Each facet is defined by vertices and can be evaluated separately using barycentric coordinates, allowing the complex polygon filtering operation to be divided into simpler, parallelizable triangular computations that reduce overall computational complexity
Solution Approach 2:
The patent replaces the traditional mechanical approach of repeatedly summing pixel values within polygon regions with a mathematical substitution using barycentric coordinate systems and transfer functions. This substitution transforms the computational problem from iterative numerical summation to direct mathematical evaluation, significantly reducing CPU operations while maintaining processing accuracy
2Reliability
If traditional filtering techniques are used for simulating camera defocus and depth of field, then image processing effects can be achieved, but excessive CPU time is consumed due to repetitive summing operations
Solution Approach 1:
The patent performs preliminary action by pre-defining the polygon kernel geometry with vertex coordinates and facet structures before the actual filtering operation. The transfer function and barycentric coordinate systems are established in advance, allowing the filtering process to directly evaluate pixel contributions without performing repetitive summing operations during execution, thus reducing CPU time significantly
Solution Approach 2:
The patent substitutes the time-consuming mechanical summation process with a mathematical evaluation system based on barycentric coordinates and transfer functions. This substitution allows the system to compute pixel contributions through direct mathematical formulas rather than iterative addition, dramatically reducing the CPU time required for each filtering operation
3Adaptability or versatility
If scaled polygon kernels are used for image processing, then different blur effects can be achieved, but computational complexity increases with kernel size
Solution Approach 1:
The patent creates a universal processing framework using barycentric coordinate systems and transfer functions that can handle polygon kernels of any scale and shape. The same mathematical apparatus evaluates facets regardless of their size, allowing the system to process small and large polygons with identical computational complexity, thus achieving scale independence and eliminating the direct relationship between kernel size and computational burden
Data Source
AI summary
A computer-implemented method includes selecting a polygon that includes a convex shape defined by vertices represented by integer coordinates. The method also includes determining a polygon transfer function from cones defined by some of the vertices. The polygon transfer function defines image processing operations independent of the scale of the polygon.