Image Hidden Information Detector Using LSB Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data leak prevention solutions are inadequate in detecting hidden information within image files, as they cannot effectively identify and differentiate between visible images, ASCII codes, binary data, and encrypted data embedded in the least significant bits of pixel values, making it difficult to determine the type of information hidden in such files.
Innovation Solution
A system and method for analyzing image files by extracting the least significant bits of pixel values, applying masks to form new pixel values, and using image/no-image detectors and convolutional neural networks to classify whether the hidden information is an image, ASCII codes, or encrypted data, thereby determining the type of information encoded within the image files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current data leak prevention solutions are used to inspect image files, then metadata fields can be cleaned of sensitive information, but hidden information embedded in pixel values cannot be detected
Solution Approach 1:
The patent extracts the least significant bits (LSBs) of pixel values from image files to isolate and analyze hidden information. By separating the LSB components from the full pixel data, the system can specifically detect concealed information without being overwhelmed by the complete image data, thereby enabling detection of hidden content that traditional metadata cleaning approaches miss.
Solution Approach 2:
The patent introduces machine learning models (image/no-image detector and classification models) as intermediary components between the raw pixel data and the final detection result. These intermediaries process the extracted LSB patterns and classify them into specific categories (image, ASCII, binary, encrypted), bridging the gap between raw data inspection and meaningful threat identification.
2Measurement precision
If the least significant bits of all pixels are extracted and analyzed, then hidden information can be detected, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the analysis process into distinct stages: first extracting LSBs from pixels, then applying an image/no-image detector to filter out non-hidden patterns, and finally using specialized classification models only on suspicious cases. This segmentation reduces the overall computational burden by avoiding full classification of every pixel in every image file.
Solution Approach 2:
The patent applies a two-stage detection approach where a quick image/no-image detector first filters the data, and only potentially suspicious cases proceed to more computationally intensive classification. This partial application of full analysis only where needed balances detection accuracy with processing efficiency.
3Measurement precision
If multiple classification models are used to identify different types of hidden information, then detection accuracy improves, but the system complexity and resource requirements increase
Solution Approach 1:
The patent implements a dynamic, multi-stage classification system where the level of analysis adapts based on the detection needs. The system starts with a lightweight image/no-image detector and progressively engages more sophisticated classification models only when necessary, allowing the system complexity to dynamically adjust rather than remaining statically high.
Solution Approach 2:
The patent replaces traditional rule-based detection mechanisms with machine learning-based classification models. These models automatically learn patterns distinguishing different types of hidden information (images, ASCII, binary, encrypted) without requiring manual programming of detection rules, thereby improving accuracy while managing complexity through automated pattern recognition.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 3
AI summary
A hidden information detector for image files extracts N least significant bits from each of a first set of pixels of an image file, wherein N is an integer greater than or equal to 1. The detector then applies a mask to each of the extracted N least significant bits to form a second set of pixel values and determines a first probability as to whether the second set of pixels encodes a hidden image. Responsive to the first probability exceeding a first threshold, the detector determines a second probability as to whether the second set of pixels matches an image encoded in the first set of pixels. Responsive to a determination that the second probability is less than a second threshold, the detector performs a non-image classifier on the second set of pixels.