Structural Feature Sequencing for Off-line Character Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing character recognition methods, particularly OCR, face challenges in recognizing characters with various fonts in natural environments due to difficulties in training for font variations, leading to low recognition rates.
Innovation Solution
A method that preprocesses character information from images using structural character recognition, detecting structural feature points like end and divergence points, assigning corresponding codes, and sequencing them to generate a numeral string, which normalizes and weights the features to enhance recognition across different fonts and angles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a statistic recognition method using a neural network is used for OCR, then high recognition rate is achieved when font is uniform or limited, but recognition rate deteriorates when characters are from natural environment with various fonts
Solution Approach 1:
The patent segments the character recognition process into distinct phases: preprocessing (binarization, thinning), normalization (rotation correction, scaling), and structural feature extraction (endpoint detection, chain code generation). This segmentation allows each phase to optimize for its specific function, enabling the system to handle various fonts effectively while maintaining high recognition accuracy.
Solution Approach 2:
The patent changes key parameters during processing: it normalizes character size and orientation, converts grayscale to binary, and transforms spatial coordinates into sequential chain codes. These parameter transformations make the recognition system invariant to font variations, scaling, and rotation, thereby improving both recognition rate and adaptability to various fonts.
2Measurement precision
If structural character recognition method is used for online input, then recognition accuracy is improved due to uniform stroke structures, but it cannot be applied to off-line character recognition from images
Solution Approach 1:
The patent applies preliminary actions to off-line character images to make them suitable for structural recognition: binarization converts the image to black and white, thinning reduces strokes to uniform width, and normalization corrects orientation. These preliminary processing steps transform raw images into a format that preserves stroke structure information, enabling structural recognition methods to be effectively applied to off-line character recognition.
Solution Approach 2:
The patent introduces an intermediary processing stage between image input and structural analysis. The preprocessing module (binarization, thinning, normalization) acts as an intermediary that converts diverse image formats into a standardized representation, bridging the gap between off-line image recognition and online structural recognition methods.
3Productivity
If character information is processed without normalization, then processing speed is maintained, but recognition accuracy deteriorates due to variations in font size and angle
Solution Approach 1:
The patent performs normalization as a preliminary action before structural feature extraction. By correcting orientation and scaling characters to a standard size beforehand, the system ensures that subsequent processing operates on uniform data, improving recognition accuracy without significantly impacting processing speed due to the efficient implementation of these preprocessing steps.
Data Source
AI summary
Disclosed is a method of sequencing character information in order to increase precision of character recognition. The method includes: a pre-processing that extracts character information from an image to binarize the extracted character information through a predetermined threshold and extracts and thins a center line of the binarized character information; normalizing the pre-processed character information to character information according to a predetermined criteria; and sequencing the normalized character information using structural features including an end point or a divergence point of the character information. The present invention suggests an angle normalization method of input character information, a structural feature position determining method, and a structural feature numeral string generating method to strongly recognize characters configured by various fonts obtained from a natural scene regardless of an angle or a size of the characters.