Parallel Coherent Random Walk Texture Interpolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing texture by numbers (TBN) techniques require artists to manually prepare input exemplars with all desired texture patterns and transitions, making them cumbersome and less adopted due to the burden of manual preparation, and conventional texture interpolation methods are either computationally intensive or unsuitable for real-time use on arbitrary 3D surfaces.
Innovation Solution
A method utilizing parallel coherent random walk searching to automatically generate high-quality transitions between texture patterns by merging first and second pattern exemplars based on a blend map, enabling real-time synthesis of composite images and transitions on arbitrary 3D surfaces, thereby automating input exemplar preparation and supplementing or replacing TBN processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual preparation of input exemplars is performed to ensure all transitions among patterns are present, then texture composition quality is improved, but the complexity and time required for preparation increases significantly
Solution Approach 1:
The system performs preliminary action by automatically generating all necessary pattern transitions during runtime based on the blend map, eliminating the need for manual pre-preparation of transition examples. The correspondence search process with blended search neighborhoods pre-computes and generates transitions on-demand as textures are synthesized.
Solution Approach 2:
The system enables self-service by allowing the texture synthesis process to automatically generate its own transition patterns during runtime. The correspondence search process with blended search neighborhoods enables the system to self-generate appropriate transitions based on the blend map without requiring external manual preparation of exemplar transitions.
2Extent of automation
If conventional texture interpolation methods are used to generate transitions automatically, then manual preparation burden is reduced, but computational intensity increases making real-time use impractical
Solution Approach 1:
The system segments the computational workload by dividing the texture synthesis into independent pixel/texel processing units that can be computed in parallel. Each pixel/texel is synthesized independently using its own correspondence search with blended search neighborhoods, enabling efficient parallel computation on GPUs and other parallel processing architectures.
Solution Approach 2:
The system changes parameters by using blended search neighborhoods that dynamically adjust the search parameters based on the blend map values. This allows the correspondence search to adapt its behavior to the local blending requirements, improving both quality and efficiency by focusing searches in relevant regions of the pattern exemplars.
3Extent of automation
If conventional texture interpolation methods are used, then automatic transition generation is achieved, but performance on arbitrary 3D surfaces is poor or unsuitable
Solution Approach 1:
The system achieves universality by designing the correspondence search process with blended search neighborhoods to work with arbitrary pattern exemplars and blend maps regardless of the underlying surface geometry. The method can synthesize textures on 2D images, 3D surfaces, and other geometries by adapting to the local neighborhood requirements of each surface point.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
There is provided a method and system for fast pattern interpolation with parallel coherent random walk searching. The system includes a system processor and a system memory, and an image processing application stored in the system memory. The image processing application, under control of the system processor, is configured to receive first and second pattern exemplars and a blend map for merging the first and second pattern exemplars, and to utilize a correspondence search process to determine first and second target images for use in producing a composite image. The first and second target images correspond respectively to the first and second pattern exemplars. In addition, a constraint imposed on the correspondence search process by the image processing application is based on the blend map for merging the first and second pattern exemplars.