Character Recognition via Gray Value Gradient Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing character recognition methods, particularly in the OCR-B font used in identification documents, struggle to accurately distinguish between the letter 'O' and the number '0' due to similarities in gray value-based methods, leading to high incorrect recognition rates, especially in context-based methods when alphanumeric serial numbers are introduced.
Innovation Solution
The method evaluates the frequencies of gray value gradient directions along contour sections, determining a ratio of horizontal to vertical or diagonal gradients to unambiguously recognize characters, eliminating the need for context information and reducing false detection rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If gray value-based character recognition methods are used, then the recognition process is simple and fast, but the ability to distinguish between similar characters (O and 0) is poor
Solution Approach 1:
The patent transforms the recognition approach by changing from using absolute gray value intensities to using gray value gradients (derivatives). This parameter transformation enhances the differentiation between O and 0 by capturing the directional changes in intensity along contour sections, allowing simple and fast recognition while improving character distinction accuracy.
Solution Approach 2:
The patent introduces a new dimension of analysis by evaluating gray value gradients not just in magnitude but also in direction (angular ranges). This dimensional expansion from scalar gray values to vector-like gradient information enables better character discrimination while maintaining computational efficiency.
2Measurement precision
If context-based character recognition methods are used, then the recognition accuracy improves for traditional MRZ, but the methods become ineffective when alphanumeric serial numbers are introduced
Solution Approach 1:
The patent enables characters to be self-identified through their intrinsic gradient properties rather than relying on external context information. Each character's contour sections generate unique gradient frequency patterns that serve as self-contained identification features, making the method adaptable to both traditional MRZ and alphanumeric serial numbers without requiring contextual validation.
3Measurement precision
If the frequency of gray value gradient directions along contour sections is evaluated, then clear recognition of O and 0 is achieved without context information, but the processing complexity increases
Solution Approach 1:
The patent segments the character contour into multiple sections and evaluates gradient frequencies independently for each section. This segmentation allows the complex evaluation to be broken down into manageable parts, processing different contour sections separately and combining results, thereby reducing overall processing complexity while maintaining high recognition accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for identifying a character that has a character contour formed by gray value curves, having the steps of determining (101) a plurality of gray value gradients using the gray value curves, determining (103) a first number of gray value gradients of the plurality of gray value gradients, said gray value gradients running within a first angular range with respect to a first direction, determining (105) a second number of gray value gradients of the plurality of gray value gradients, said gray value gradients running within a second angular range with respect to a second direction, and identifying (107) the character using a relation between the first number of gray value gradients and the second number of gray value gradients.