Digital Ink Stroke Grouping With Heuristic Confidence Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for grouping digital ink strokes are either computationally expensive or inaccurate, particularly when dealing with different languages and handwriting sizes, leading to over-grouping or under-grouping issues.
Innovation Solution
A heuristic-based approach that computes group and stroke heuristics to determine confidence values for grouping digital ink strokes, without relying on computationally expensive ink analysis or pure proximity-based models, using machine learning to refine weights and thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML-based ink analysis models are used to recognize words and set up groups of strokes, then grouping accuracy is improved, but processing time increases to 500 ms or more and resource consumption increases
Solution Approach 1:
The patent segments the complex ML-based ink analysis into multiple simpler heuristic components: proximity heuristic, orientation heuristic, temporal heuristic, and size heuristic. Each heuristic evaluates a specific aspect of stroke relationships independently, and their results are combined to determine grouping. This segmentation maintains grouping accuracy while significantly reducing processing time by avoiding a single complex ML model.
Solution Approach 2:
The patent applies partial action by using only the necessary subset of heuristics for each grouping decision rather than applying a comprehensive ML model to all strokes. The system evaluates strokes against multiple heuristics but only applies the relevant ones based on the specific context, reducing overall computational overhead while maintaining accuracy.
2Productivity
If pure proximity-based models are used with hard-coded distance thresholds, then processing speed is improved, but grouping accuracy deteriorates causing over-grouping or under-grouping
Solution Approach 1:
The patent changes the parameters used for grouping from simple proximity distance to a multi-dimensional heuristic evaluation including proximity, orientation, temporal relationships, and size. Instead of using a single hard-coded distance threshold, the system dynamically adjusts grouping decisions based on multiple varying parameters, which improves accuracy while maintaining processing speed through efficient heuristic calculations.
Solution Approach 2:
The patent introduces dynamics by making the grouping criteria adaptive rather than static. The heuristic weights and thresholds are not fixed but can be adjusted based on the specific stroke characteristics and context. This dynamic approach allows the system to accurately handle different writing styles, languages, and scenarios without sacrificing processing speed.
3Adaptability or versatility
If ML models are trained for each language to recognize words and group strokes, then adaptability to different languages is improved, but device complexity and resource requirements increase
Solution Approach 1:
The patent creates a universal heuristic-based grouping system that works across multiple languages without requiring separate ML models for each language. The heuristics evaluate fundamental stroke relationships (proximity, orientation, timing, size) that are consistent across different writing systems. This multi-functional approach allows the same grouping mechanism to handle various languages and writing styles, reducing device complexity while maintaining adaptability.
Data Source
Figure 1
Figure 2
Figure 3A~3D
AI summary
Technologies are disclosed for grouping strokes of digital ink. The disclosed technologies receive a new stroke of digital ink and, responsive thereto, identify groups of strokes that are candidates for receiving the new stroke. Heuristics are computed for one of the candidate groups and for the new stroke. A confidence value is computed for the new stroke based upon the heuristics for the new stroke and the candidate group. If the confidence value exceeds a confidence threshold, which can be user-adjustable, the new stroke of digital ink is added to the candidate group. This process can then be repeated for other candidate groups until a group is found for the new stroke. If no group is found, a new group can be created for the new stroke. A machine learning model can also generate weights for use in computing confidence values.