Vector Text Extraction Using OCR-Guided Path Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vector-based image editing systems are limited in their ability to extract vector text efficiently and accurately from complex vector images, requiring tedious cleanup processes due to the inability to distinguish between vector paths depicting text and non-text content.
Innovation Solution
A multistep approach involving content analysis, candidate outline filtering, and conditional candidate outline pruning using an optical character recognition model to identify and extract textual content, determine bounding boxes, and apply constraints to filter and refine vector paths, allowing for minimal user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional vector path selection methods are used to extract text from vector images, then the system can identify vector paths, but the accuracy is poor and requires tedious cleanup processes
Solution Approach 1:
The patent introduces an optical character recognition (OCR) model as an intermediary component that analyzes vector paths and identifies which ones represent text. This mediator processes the vector paths and provides accurate text identification, eliminating the need for manual cleanup processes while maintaining high extraction accuracy
Solution Approach 2:
The patent replaces the manual mechanical process of selecting and cleaning up vector paths with an automated OCR-based system. The OCR model automatically analyzes vector path characteristics and identifies text-containing paths, substituting the tedious manual cleanup process with an intelligent automated solution
2Ease of operation
If manual selection of vector paths is used, then the system can extract text, but the ease of operation is poor due to excessive user interactions required
Solution Approach 1:
The patent implements a self-service mechanism where the OCR model automatically performs text extraction without requiring user interactions for path selection or cleanup. The system serves itself by autonomously identifying text vector paths and extracting the text, eliminating the need for manual intervention and significantly improving both ease of operation and productivity
3Manufacturing precision
If simple vector path filtering is used, then the processing speed is fast, but the manufacturing precision is poor in distinguishing text from non-text paths
Solution Approach 1:
The patent introduces an OCR model as an intermediary that bridges the gap between simple vector path filtering and accurate text identification. This intermediary component adds intelligence to the filtering process, enabling high precision in distinguishing text from non-text paths while maintaining a manageable system architecture
Solution Approach 2:
The patent makes the OCR model serve multiple functions: it not only identifies text-containing vector paths but also provides bounding box information and helps filter non-text paths. This multi-functionality increases precision without proportionally increasing system complexity, as the same OCR component performs multiple tasks
Data Source
AI summary
The present disclosure is directed toward systems, methods, and non-transitory computer readable media that that extract vector text from vector images using a multistep approach that involves content analysis, candidate outline filtering, and conditional candidate outline pruning. In particular, in one or more embodiments, the disclosed systems utilize an optical character recognition model to extract textual content as well as bounding boxes corresponding to the textual content from within vector images. The disclosed systems determine a set of intersecting vector paths that overlap the bounding boxes corresponding to the textual content. The disclosed systems apply various constraints to the set of intersecting paths to filter the paths and determine a set of text vector paths that outlines the textual content.