Dynamic Upscaling for Sub-Pixel Digital Image Vectorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems face inefficiencies, inaccuracies, and operational inflexibilities in vectorizing raster images due to computational costs, limited GPU/CPU resources, and domain gaps, particularly on mobile devices, and struggle with accurately performing image vectorization beyond natural images.
Innovation Solution
A selective super-resolution system that identifies high-frequency portions of a raster image using edge detection, applies an image super-resolution model selectively to these portions, and combines them with upscaled low-frequency segments to generate a segmentation map for vectorization, trained on an augmented dataset of synthetically modified images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional vectorization models process entire raster images, then vectorization quality is maintained, but computational efficiency deteriorates and processing time increases
Solution Approach 1:
The patent divides the raster image into multiple patches and processes only the high-frequency regions (edges, textures) through super-resolution while applying simpler upscaling to low-frequency regions. This segmentation strategy reduces computational load by applying heavy processing only where needed, thereby improving productivity without sacrificing overall vectorization quality.
Solution Approach 2:
The patent applies different processing quality to different regions of the image. High-frequency regions receive sophisticated super-resolution treatment to preserve critical details, while low-frequency regions receive simpler upscaling. This local quality differentiation optimizes computational efficiency by concentrating resources on regions that most impact vectorization accuracy.
2Measurement precision
If image super-resolution is applied to entire images, then vectorization accuracy is improved, but device resource requirements increase and mobile device compatibility deteriorates
Solution Approach 1:
By segmenting the image into patches and identifying only high-frequency regions for super-resolution processing, the patent significantly reduces the computational footprint. This allows the system to maintain high vectorization accuracy for critical regions while reducing overall resource consumption to levels compatible with mobile devices.
Solution Approach 2:
Instead of applying super-resolution to the entire image (excessive action), the patent applies it selectively only to high-frequency regions where it is most needed (partial action). This partial application maintains sufficient vectorization accuracy while dramatically reducing GPU/CPU resource requirements, enabling deployment on mobile platforms.
3Speed
If conventional super-resolution models are used, then processing speed is maintained, but accuracy deteriorates for non-natural images due to domain gaps
Solution Approach 1:
The patent modifies the training parameters and architecture of the super-resolution model to accommodate diverse image domains including non-natural images. By adjusting model parameters and using domain-adaptive training, the system maintains processing speed while improving accuracy across multiple domains, eliminating the domain gap limitation of conventional models.
Data Source
AI summary
The present disclosure relates to systems, methods, and non-transitory computer-readable media that selectively utilizes an image super-resolution model to upscale image patches corresponding to high frequency portions. In particular, the disclosed systems select a set of image patches corresponding to high frequency portions of a digital image at a first resolution. Furthermore, the disclosed systems utilize an image super-resolution model to generate upscaled image patches for the set of image patches of the high-frequency portions to a second resolution higher than the first resolution according to an upscaling factor of at least two. The disclosed systems generate a segmentation map of the digital image based on the upscaled image patches and an upscaled segmentation corresponding to low-frequency portions of the digital image. Further, the disclosed systems generate a vectorized digital image for the digital image according to the segmentation map.