Single Stroke Character Recognition Using Covariance Adjusted Distances
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Handwriting recognition systems face challenges in accurately identifying single stroke characters due to variations in user input, as existing methods fail to effectively account for correlations between coordinate parameters, leading to reduced robustness and accuracy in character classification.
Innovation Solution
A system that normalizes input trajectories to a standard size, generates equidistant reference points, and uses covariance adjusted distances to compare them with class templates, accounting for correlations between coordinates through the Mahalanobis distance, thereby improving discrimination among single stroke characters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional distance metrics are used to compare trajectories with class templates, then the computation is simple, but the recognition accuracy decreases due to inability to account for coordinate correlations
Solution Approach 1:
The patent transforms the distance calculation by incorporating covariance information into the metric itself. Instead of using simple Euclidean distance, the system uses Mahalanobis distance which accounts for coordinate correlations through the covariance matrix. This parameter change in the distance metric enables the system to handle variations in user input more effectively while maintaining computational feasibility through pre-computed template statistics.
2Reliability
If the system accounts for coordinate correlations using covariance adjusted distances, then the robustness to user input variations improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary computation of mean values and covariance matrices for each character class template before actual recognition occurs. These pre-computed statistics are stored and reused during recognition, eliminating the need to recalculate them for each input. This preliminary action shifts computational burden from the recognition phase to the template preparation phase, improving real-time robustness without proportionally increasing ongoing computational complexity.
Solution Approach 2:
The system changes the distance metric from simple Euclidean to Mahalanobis distance, which incorporates covariance information. This parameter change in the measurement approach enables the system to naturally handle coordinate correlations and user input variations. The Mahalanobis distance formula automatically adjusts for the statistical relationships between coordinates, providing robustness without requiring complex adaptive algorithms during recognition.
3Measurement precision
If more detailed trajectory data is used for recognition, then the discrimination ability improves, but the data processing requirements and complexity increase
Solution Approach 1:
The patent extracts only the essential statistical features from the trajectory data - specifically the mean coordinates and covariance values for each reference point. Instead of processing the entire continuous trajectory or all individual data points, the system extracts these condensed statistical representations. This extraction reduces the data processing load significantly while retaining the discriminative information needed for accurate character recognition.
Solution Approach 2:
The system transforms the raw trajectory data into statistical parameters (means and covariances) for each reference point. This parameter transformation converts a large volume of continuous trajectory data into a compact set of statistical descriptors. The mean-covariance representation captures the essential characteristics of each character class while reducing the data dimensionality and processing requirements for recognition.
Data Source
AI summary
Systems and methods are provided for recognizing a single stroke gesture. An input trajectory representing the single stroke gesture is received. The input trajectory is normalized to produce a normalized trajectory having a standard set of dimensions. The normalized trajectory is reduced to a standard number of substantially evenly spaced points. Respective covariance adjusted distances are determined for each class as the covariance adjusted position between a set of points representing each single stroke character classes and the substantially evenly spaced points representing the normalized trajectory.