Bit Sequence Fingerprinting for Spam Image Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating fingerprints of bit sequences, particularly in email processing environments, are inefficient due to high computational costs and susceptibility to misclassification, especially when dealing with spam emails containing embedded images with random variations and distortions.
Innovation Solution
A method that determines the relative occurrence frequency of bit combinations in a bit sequence, calculates a difference value compared to random occurrence frequency, and assigns these values to bins to generate a fingerprint, which can identify similar bit sequences even with slight variations, using a computer system to quickly process the bit sequence and reduce computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If OCR techniques and image analysis methods are used to extract features from embedded images, then the ability to identify spam emails with embedded images improves, but the computational cost and processing time increase significantly
Solution Approach 1:
The patent segments the image processing task by dividing the image into multiple blocks and extracting features from each block independently. This segmentation allows parallel processing of different regions, reducing overall computational complexity while maintaining the ability to detect spam indicators in any part of the image.
Solution Approach 2:
The patent extracts only the essential spatial and color features from image blocks rather than performing full OCR analysis. By taking out only the necessary features (spatial relationships and color distributions) needed for spam detection, the system achieves reliable identification without the high computational cost of complete character recognition.
2Productivity
If traditional hash functions are applied to embedded images, then the process is computationally efficient, but the ability to identify spam emails with distorted or varied images deteriorates
Solution Approach 1:
The patent applies local quality analysis by examining specific regions (blocks) of the image rather than treating the entire image uniformly. Each block is analyzed for local spatial and color characteristics, allowing the system to identify spam patterns even when parts of the image are distorted or vary randomly, while maintaining efficient processing through localized rather than global analysis.
3Measurement precision
If sophisticated image analysis techniques are used to extract unique descriptors, then the precision of image fingerprinting improves, but the computational complexity and processing cost increase
Solution Approach 1:
The patent changes the parameters being measured from complex visual features to simpler spatial and color statistics. By measuring block positions, color distributions, and spatial relationships rather than attempting to recognize text or complex image features, the system achieves sufficient fingerprint uniqueness for spam detection while dramatically reducing computational complexity and avoiding the need for sophisticated image analysis infrastructure.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
There is provided a method of generating a fingerprint of a bit sequence comprising: determining a relative occurrence frequency of each bit combination of a set of bit combinations in the bit sequence, wherein the set of bit combinations comprises all possible non-redundant sub-sequences of bits having at least one bit and at most a preset maximal number of bits; determining for each bit combination of the set of bit combinations a difference value between the relative occurrence frequency of the bit combination and a random occurrence frequency, the random occurrence frequency relating to the expected random occurrence of the bit combination in the bit sequence; allocating a set of bins, each bin of the set of bins being associated with a pregiven interval of difference values, each bin further relating to a bin value; assigning the difference value of each bit combination to the bin which is associated with the interval of difference values in which the difference value of the corresponding bit combination lies; generating the fingerprint of the bit sequence by use of the bin values of the bins to which a difference value has been assigned.