Obfuscated Text Deciphering via Image Recognition and Similarity Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to accurately recognize and decipher obfuscated text, such as Leet words, in unstructured cyber security documents, hindering effective natural language processing.
Innovation Solution
A method and apparatus that converts text into an image, uses a text recognition model to identify obfuscated characters, and deciphers them through a database comparison with stored obfuscated character strings, utilizing similarity measures and a character string search tree to determine and decode the obfuscated text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If obfuscated text is converted to image and recognized using text recognition model, then recognition accuracy of obfuscated text is improved, but processing complexity and time increase
Solution Approach 1:
The system segments the obfuscated text processing into distinct stages: image conversion, text recognition, similarity calculation, and database matching. Each stage handles a specific aspect of the problem, allowing the complex task to be managed through modular components that can be optimized independently.
Solution Approach 2:
The patent introduces an intermediate text recognition model that converts obfuscated text into images and then back into readable text. This intermediary process acts as a bridge between the obfuscated input and the deciphering system, enabling accurate recognition without requiring direct analysis of the obfuscated characters.
2Measurement precision
If similarity comparison is performed between target character string and recognized character string, then obfuscation detection accuracy is improved, but computational time increases
Solution Approach 1:
The system performs preliminary actions by pre-processing the text into images and using the text recognition model to generate candidate deciphered text before the similarity comparison stage. This preparation work is done in advance to reduce the computational burden during the actual obfuscation detection phase.
Solution Approach 2:
The patent changes the parameter space by transforming the text from character-based representation to image-based representation, and then to a similarity metric space. This parameter transformation allows for more accurate obfuscation detection while managing computational complexity through efficient similarity calculations.
3Measurement precision
If database search is performed to decipher obfuscated text, then deciphering accuracy is improved, but search time and memory usage increase
Solution Approach 1:
The system applies local quality by focusing the database search on specific candidate strings generated by the text recognition model, rather than searching the entire database. This localized approach maintains high deciphering accuracy by concentrating computational resources on the most promising candidates.
Solution Approach 2:
The patent uses partial action by performing database search only on the subset of candidate strings that have high similarity scores, rather than exhaustively searching all possible decryptions. This selective approach achieves sufficient deciphering accuracy while significantly reducing search time and memory usage.
Data Source
AI summary
Provided is a method for deciphering obfuscated text for cyber security and an apparatus for the same. The method according to some embodiments includes: converting text including a target character string into an image; recognizing a character string in the image using a text recognition model; and determining that the target character string is an obfuscated character string, based on a similarity between the target character string and the recognized character string being equal to or less than a first reference value.