Ink Element Recognition via Perimeter-Based Stroke Construction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OCR and ICR/IWR systems face challenges in accurately recognizing text characters due to noise and natural variations in handwritten content, leading to incorrect skeletonization and recognition issues.
Innovation Solution
The system determines the perimeter of an ink element, identifies pen units based on dimension and position characteristics, and constructs strokes to recognize objects, using iterative methods to best fit the ink element's contour, thereby reducing noise-induced errors and accurately replicating the original pen trajectory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional skeletonization methods are used to extract character skeletons from imaged text, then the process is simple and fast, but noise in the captured images causes spurious artefacts in the skeletons leading to inaccurate character recognition
Solution Approach 1:
The patent segments the skeletonization process into multiple distinct stages: (1) generating initial skeleton pixels using thinning algorithms, (2) identifying and removing spurious branches through iterative pruning, (3) reconstructing the cleaned skeleton. This segmentation allows each stage to address specific problems independently, improving overall accuracy without requiring a completely complex new approach.
Solution Approach 2:
The patent implements feedback mechanisms where the skeletonization process iteratively refines the skeleton by: (1) generating an initial skeleton, (2) detecting spurious branches through analysis of pixel connectivity and geometry, (3) removing identified artefacts, and (4) repeating the process until convergence. This feedback loop continuously improves skeleton quality by eliminating noise-induced errors.
2Measurement precision
If iterative skeletonization methods are used to improve accuracy by removing spurious artefacts, then character recognition accuracy improves, but processing time increases
Solution Approach 1:
The patent applies partial action by implementing conditional stopping criteria in the iterative skeletonization process. The algorithm performs multiple refinement passes but terminates early when: (1) a maximum number of iterations is reached, or (2) no spurious branches are detected in consecutive passes, or (3) the skeleton stabilizes. This prevents excessive processing while ensuring sufficient cleaning of artefacts.
Solution Approach 2:
The patent applies local quality by focusing computational effort only on regions of the skeleton that contain spurious artefacts rather than uniformly processing the entire skeleton. The algorithm identifies problematic areas through local geometric analysis of pixel connectivity and branch characteristics, then applies cleaning operations only where needed, reducing overall processing time.
3Productivity
If simple thinning methods are used for skeletonization, then processing is fast and computationally efficient, but the skeletons contain branches and segments that are not part of the intended character
Solution Approach 1:
The patent applies preliminary action by performing initial skeleton generation using fast thinning algorithms before applying refinement operations. The process sequence is: (1) generate initial skeleton quickly using parallel thinning, (2) identify spurious branches in the initial skeleton, (3) remove artefacts, and (4) produce final cleaned skeleton. This preliminary fast processing maintains high productivity while subsequent targeted refinement improves reliability.
Data Source
AI summary
A system for recognition of objects from ink elements on a computing device is provided. The computing device comprises a processor, a memory and at least one non-transitory computer readable medium for recognizing content under control of the processor. The at least one non-transitory computer readable medium is configured to determine a perimeter of an ink element stored in a memory of the computing device, determine a plurality of pen units for the ink element based on the determined ink element perimeter, determine at least one stroke representing a path through two or more of the pen units, and cause recognition of one or more objects represented by the ink element using the determined at least one stroke.