OCR String Identification via Rank-Based Cost Minimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical Character Recognition (OCR) systems face challenges in accurately identifying multiple strings of characters in images due to non-ideal acquisition conditions, partial occlusion, and structural similarities between characters, leading to errors in recognizing vehicle license plate numbers, serial numbers, and labels on packages, especially with the presence of noise and additional text that can be erroneously identified as the searched text.
Innovation Solution
The system employs a method to accurately identify multiple strings by receiving a set of ranks indicative of string positions, performing optical character recognition, determining minimum assignation costs between candidate strings and model strings, and selecting a subset with the lowest cost, ensuring compatibility and adherence to defined model string characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR is used to identify strings in images, then the process is simple and fast, but accuracy deteriorates due to noise, occlusion, and structural similarity between characters
Solution Approach 1:
The patent segments the string identification process into multiple independent components: candidate string extraction, model string definition, rank assignment, and cost calculation. Each component handles a specific aspect of the identification task, allowing the system to process complex images by breaking them down into manageable segments that can be evaluated independently and combined to achieve high accuracy.
Solution Approach 2:
The patent performs preliminary actions by defining model strings with expected characteristics (length, character patterns, position ranges) before actual identification occurs. This pre-definition creates a framework that guides the subsequent identification process, allowing the system to filter and evaluate candidate strings against known criteria, thereby improving accuracy while managing complexity through structured preparation.
2Reliability
If OCR identifies all text in an image, then comprehensive text detection is achieved, but error rate increases due to identification of irrelevant text
Solution Approach 1:
The patent applies local quality by defining specific characteristics for model strings at different positions (ranks) in the image. Each rank has associated model strings with particular expected properties (length, character composition, position constraints). This localized quality control allows the system to distinguish relevant text from irrelevant text by matching local characteristics against predefined models, thereby improving reliability while reducing false positives.
Solution Approach 2:
The patent implements feedback through the cost calculation mechanism that evaluates candidate strings against model strings. The system calculates costs based on mismatches between candidate and model characteristics, then uses this feedback to select the best matching strings. This feedback loop continuously refines the identification by comparing results against expected patterns and adjusting selections to minimize errors, enhancing reliability while filtering out false positives.
3Measurement precision
If multiple candidate strings are evaluated, then identification thoroughness is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by evaluating only the most promising candidate strings against model strings rather than exhaustively analyzing all possible text regions. The system extracts candidate strings, defines model strings with specific criteria, and evaluates matches based on cost calculations. This partial evaluation approach focuses computational resources on the most likely candidates, maintaining high identification accuracy while significantly reducing processing time compared to exhaustive methods.
Data Source
AI summary
A method and system of identifying a plurality of strings in an image are described. A set of ranks and an image are received. Optical character recognition is performed on the image to identify candidate strings. The candidate strings are ordered according to their location in the image with respect to a reading direction. Minimum assignation costs between the plurality of candidate strings and the set of ranks are determined. Each of the minimum assignation costs is representative of a cost of one or more edit operations performed on the plurality of candidate strings to obtain a subset of candidate strings ordered according to the set of ranks. A subset of candidate strings that is associated with the lowest minimum assignation cost when compared with other minimum assignation costs of other subsets of candidate strings from the plurality of candidate strings is selected and outputting.