Rotation-Free Handwritten Character Recognition via Stroke Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing character recognition systems are highly dependent on the orientation of input characters, leading to reduced accuracy when characters are not received in a predefined orientation, which is a common issue when inputting characters via mobile devices where the orientation may vary.
Innovation Solution
The system rotates incoming characters to a normalized orientation based on the relationship between key points of their strokes before recognition, allowing for orientation-free recognition by pre-processing training and input characters to align them in a consistent direction, and constructing a recognition model that is invariant to character orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If character recognition is performed based on predefined orientation, then recognition accuracy is improved for upright characters, but recognition accuracy deteriorates sharply when characters deviate from predefined orientation
Solution Approach 1:
The patent applies preliminary action by rotating input characters to a normalized orientation before performing recognition. The system calculates the orientation angle of each character and pre-rotates it to align with the predefined upright position, ensuring that the recognition model receives characters in the expected orientation. This preliminary orientation correction resolves the contradiction by preparing characters in advance to match the recognition model's requirements.
Solution Approach 2:
The patent employs parameter changes by transforming the orientation parameter of input characters. The system detects the actual orientation angle of each character and applies a rotation transformation to change its orientation parameter from an arbitrary angle to a standardized upright position. This parameter transformation allows the recognition system to maintain high accuracy regardless of the original input orientation.
2Adaptability or versatility
If rotation processing is applied to normalize character orientation, then orientation adaptability is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent applies partial action by selectively processing only the orientation aspect of characters rather than performing comprehensive image processing. The system calculates orientation angles using simplified methods based on character bounding boxes or dominant orientation detection, applying only the necessary rotation to achieve normalization. This selective approach reduces processing overhead while maintaining orientation adaptability.
Solution Approach 2:
The patent substitutes complex mechanical image processing with computational geometry methods. Instead of using intensive image warping or complex transformation algorithms, the system employs straightforward angle calculation and rotation matrix operations. This substitution reduces computational complexity and processing time while achieving the same orientation normalization effect.
3Measurement precision
If feature extraction is performed after rotation, then recognition accuracy for rotated characters is improved, but processing complexity increases compared to direct feature extraction
Solution Approach 1:
The patent applies segmentation by dividing the character recognition process into distinct stages: orientation detection, rotation normalization, and feature extraction. By separating these operations, the system can optimize each stage independently. The rotation step prepares standardized input for the feature extraction module, improving accuracy without requiring complex integrated processing. This modular segmentation reduces overall system complexity.
Solution Approach 2:
The patent introduces rotation normalization as an intermediary step between input character reception and feature extraction. This intermediary process transforms characters into a standardized orientation format that simplifies subsequent feature extraction. The rotation mediator ensures that the feature extraction module receives consistently oriented input, improving recognition accuracy while keeping the feature extraction logic itself relatively simple.
Data Source
AI summary
A character recognition system receives an unknown character and recognizes the character based on a pre-trained recognition model. Prior to recognizing the character, the character recognition system may pre-process the character to rotate the character to a normalized orientation. By rotating the character to a normalized orientation in both training and recognition stages, the character recognition system releases the pre-trained recognition model from considering character prototypes in different orientations and thereby speeds up recognition of the unknown character. In one example, the character recognition system rotates the character to the normalized orientation by aligning a line between a sum of coordinates of starting points and a sum of coordinates of ending points of each stroke of the character with a normalized direction.