Video Pornography Detection via Motion Vector Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting pornographic content in video sequences are inefficient due to high false positives, inability to analyze motion, and increased processing costs, particularly when dealing with black and white images or multiple subjects, and often require decoding video signals.
Innovation Solution
A method that utilizes motion vectors within encoded video files to identify repetitive motion patterns without pixel or frequency domain analysis, reducing false positives and enabling faster detection by focusing on motion-based analysis, which can be implemented in media players and servers without decoding the video signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If skin detection techniques are used to identify adult content, then detection capability is provided, but false positives increase significantly
Solution Approach 1:
The detection process is segmented into multiple independent analysis stages: motion vector analysis, skin tone detection, and contextual pattern recognition. Each stage processes specific features separately and combines results to make final determination, reducing false positives by not relying solely on skin detection
Solution Approach 2:
Motion vector analysis serves as an intermediary layer between raw video data and adult content classification. The system first analyzes motion patterns to identify suspicious sequences, then applies skin detection only to those sequences, reducing overall false positives from skin detection alone
2Measurement precision
If video signals are decoded for analysis, then comprehensive content inspection is enabled, but processing time and computational costs increase
Solution Approach 1:
The system performs partial decoding only to the extent necessary to extract motion vectors and basic frame data. Full video decoding is avoided unless motion analysis triggers a suspicion threshold, at which point more comprehensive analysis is performed selectively on suspicious segments rather than entire videos
Solution Approach 2:
Motion vector analysis is performed preliminarily on encoded video data before any decoding occurs. This preliminary action identifies suspicious sequences that warrant further detailed analysis, filtering out normal content without requiring full decoding and thus saving significant processing time
3Productivity
If motion information is analyzed using skin detection, then motion-based detection is achieved, but processing complexity and time requirements increase
Solution Approach 1:
The system extracts only the essential motion information contained in motion vectors from the encoded video stream, rather than analyzing full pixel data or performing complex frequency domain transformations. This extraction approach provides sufficient motion-based detection capability with minimal processing complexity
Data Source
AI summary
A method for detecting the presence of pornographic contents in a sequence of video frames having associated respective motion vectors, includes identifying groups of motion vectors having similar orientation and the presence, in subsequent frames of the sequence, of motion types defined by affiliated groups of motion vectors having homologous motion characteristics in subsequent frames. Thereafter, detection in these motion types of the occurrence of negative affiliations representative of an inversion in the motion of the respective group of vectors. If the number of such negative affiliations counted at a given interval reaches a given count threshold, identified in the sequence of frames is a periodic motion susceptible of having pornographic content. Repetition of the periodic motions may be verified, the image sequence identified as susceptible to having pornographic contents and/or subjected to an optional verification procedure such as “skin detection” type, and transmission may be blocked.