Adversarial Text Detection in Images Using Shape Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying inappropriate content in images, such as spam emails, are hindered by anti-OCR features that confuse optical character recognition, making it difficult to distinguish spammy content from legitimate messages.
Innovation Solution
The approach involves selecting specific expressions and searching for similar sections within images, scoring their similarity, and determining if the expressions match, allowing for accurate identification of inappropriate content without relying solely on text extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCR techniques are used to extract text from images for spam detection, then text identification capability is improved, but anti-OCR features in spam images cause detection accuracy to deteriorate
Solution Approach 1:
Instead of extracting text from images and analyzing it (traditional OCR approach), the patent inverts the approach by embedding known spam expressions into image templates and searching for matching sections within the image. This reversal allows direct comparison with expected spam patterns while being immune to anti-OCR obfuscation techniques.
Solution Approach 2:
The patent pre-generates image templates containing spam expressions before receiving target images for analysis. By preparing reference templates in advance with known spam content, the system can directly compare these templates against sections of target images, enabling proactive detection without relying on post-hoc text extraction that may be thwarted by anti-OCR features.
2Productivity
If traditional keyword filters are used to detect spam, then detection speed is improved, but embedding inappropriate content in images causes detection accuracy to deteriorate
Solution Approach 1:
The patent creates template copies of images containing known spam expressions and uses these templates to search within target images. By copying and comparing image sections rather than extracting and analyzing text, the system maintains fast processing speeds while achieving accurate detection of image-embedded spam content that would evade traditional keyword filters.
Data Source
AI summary
An adversarial approach in detecting inappropriate text content in images. An expression from a listing of expressions may be selected. The listing of expressions may include words, phrases, or other textual content indicative of a particular type of message. Using the selected expression as a reference, the image is searched for a section that could be similar to the selected expression. The similarity between the selected expression and the section of the image may be in terms of shape. The section may be scored against the selected expression to determine how well the selected expression matches the section. The score may be used to determine whether or not the selected expression is present in the image.