Robust Low-Complexity Video Fingerprinting via Block Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video fingerprinting methods are not robust enough to correctly identify videos that have undergone severe alterations, such as encoding artifacts or noise insertion, and suffer from high computational complexity, making them unsuitable for practical applications, especially in portable devices.
Innovation Solution
A method and apparatus for robust, low-complexity video fingerprinting that involves removing frame margins, dividing frames into blocks and sub-blocks, computing block-level features as pixel means, and using a sliding window approach with a perceptual distance metric for matching, which reduces computational overhead while maintaining robustness against artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If key-frame selection algorithms are used for video fingerprinting, then identification accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent extracts only the necessary luminance component from video frames and removes frame margins to eliminate unnecessary computational overhead while retaining the essential information needed for accurate video identification
Solution Approach 2:
The patent divides video frames into blocks and further into sub-blocks, computing features at multiple levels (block-level and sub-block-level) to achieve accurate identification with reduced computational complexity compared to processing entire frames
2Reliability
If traditional video fingerprinting methods are used, then video identification is achieved, but robustness against severe alterations (encoding artifacts, noise insertion) is insufficient
Solution Approach 1:
The patent converts the harmful effect of severe alterations (noise, compression artifacts) into a benefit by using a perceptual distance metric that is specifically designed to be insensitive to these distortions, allowing accurate identification even when videos are significantly modified
Solution Approach 2:
The patent changes the parameter space by using perceptual distance metrics instead of traditional Euclidean distance, and by operating in the luminance domain rather than full color space, making the fingerprinting process inherently more robust to common video alterations
3Ease of operation
If existing video fingerprinting algorithms are implemented, then video matching is performed, but computational overhead is high making them unsuitable for portable devices
Solution Approach 1:
The patent extracts only the luminance component and removes frame margins, significantly reducing the amount of data that needs to be processed while maintaining identification accuracy, making the algorithm suitable for resource-constrained portable devices
Solution Approach 2:
The patent segments video processing into frame-level and block-level operations with early termination capability, allowing the system to stop processing once a match is found, thereby improving computational efficiency for portable device implementation
Data Source
AI summary
An apparatus and method for video fingerprinting are provided. The method includes, for each frame of a video sequence including a plurality of frames, removing a portion of the frame, dividing a remaining portion of the frame into blocks, dividing each block into sub-blocks, computing a block level feature as a mean of pixels in each sub-block within the block, concatenating all block level features in the frame, and concatenating features of all frames in the video sequence.