Adversarial OCR for Anti-OCR Spam Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing spam filtering technologies face challenges in identifying inappropriate content in images, especially when such content is embedded in a way that confuses optical character recognition (OCR) systems, known as anti-OCR features, making it difficult to distinguish spam emails from legitimate ones.
Innovation Solution
A pure adversarial OCR module is used to search for specific terms in images by splitting them into character-blocks, estimating the probability of matches, and scoring the similarity between the image blocks and the search terms, allowing for more accurate identification of inappropriate content without relying on text extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCR techniques are used to identify text in images, then text extraction capability is improved, but accuracy deteriorates when anti-OCR features are present
Solution Approach 1:
Instead of using OCR to extract text and then analyzing it (traditional approach), the patent inverts the approach by using image processing techniques to directly analyze visual characteristics of potential text regions. This inversion bypasses the vulnerability to anti-OCR features while maintaining the ability to detect text-like patterns in images.
Solution Approach 2:
The patent segments the image into character-blocks that are evaluated individually for text-like characteristics. By dividing the image into smaller regions and assessing each block's visual properties (aspect ratio, pixel density, edge patterns), the system can identify text content without relying on full OCR processing, thus avoiding anti-OCR feature exploitation.
2Productivity
If spam filtering uses keyword and statistical filters, then detection of obvious spam is improved, but detection of hidden spam in images deteriorates
Solution Approach 1:
The patent creates a multi-functional spam detection system that combines traditional keyword/statistical filtering with new image processing capabilities. The same email processing pipeline handles both text-based spam detection and image-based text detection, providing universal coverage against different spam types without requiring separate specialized systems.
Solution Approach 2:
The patent introduces an intermediary image processing layer that bridges the gap between visual content and text analysis. Instead of directly applying OCR to images (which fails against anti-OCR features), the system uses intermediate image processing steps to extract text-like pattern information, which then feeds into the spam detection logic, effectively mediating between image data and spam classification.
Data Source
AI summary
A pure adversarial optical character recognition (OCR) approach in identifying text content in images. An image and a search term are input to a pure adversarial OCR module, which searches the image for presence of the search term. The image may be extracted from an email by an email processing engine. The OCR module may split the image into several character-blocks that each has a reasonable probability of containing a character (e.g., an ASCII character). The OCR module may form a sequence of blocks that represent a candidate match to the search term and calculate the similarity of the candidate sequence to the search term. The OCR module may be configured to output whether or not the search term is found in the image and, if applicable, the location of the search term in the image.