Robust Video Signature Algorithm for Format-Invariant Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately compare and identify similar video content across different resolutions, bit rates, and container formats, leading to inefficiencies in video copy detection, indexing, and searching, as well as challenges in detecting copyright violations.
Innovation Solution
The Robust Video Signature (RVS) algorithm generates unique video signatures that are invariant to changes in resolution, bit rate, and container format, allowing for comparative analysis by calculating frame signatures, determining frame distances, and applying area augmentation to identify similarities between video files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional video comparison methods are used, then video files can be compared, but accuracy deteriorates when resolution, bit rate, or container format changes
Solution Approach 1:
The patent transforms video files into a standardized parameter space by converting them into sequences of feature vectors that are invariant to resolution, bit rate, and container format changes. This is achieved through extracting robust visual features and representing them in a normalized form that maintains semantic meaning while being independent of encoding parameters.
Solution Approach 2:
The patent uses disposable feature vectors that can be quickly generated and discarded during comparison operations. Instead of maintaining complex original video data, the system creates lightweight signature representations that enable fast, accurate comparisons without being affected by source file variations.
2Measurement precision
If frame-by-frame comparison is performed, then detailed video analysis is achieved, but computational complexity increases
Solution Approach 1:
The patent segments video content into meaningful units and represents each frame as a feature vector rather than processing raw pixel data. This segmentation approach maintains detailed analysis capability while reducing computational burden by working with compressed feature representations.
Solution Approach 2:
The patent extracts essential visual features from video frames and discards redundant information. By taking out only the discriminative features needed for comparison and eliminating unnecessary data, the system achieves high precision with reduced computational complexity.
3Measurement precision
If maximum detection is used to find video matches, then precise location is achieved, but false positives increase due to multiple local maxima
Solution Approach 1:
The patent extends the comparison from detecting single maximum similarity points to measuring the area under the similarity curve across multiple dimensions. This dimensional transformation allows the system to distinguish true matches from false positives by considering the overall shape and extent of similarity rather than isolated peaks.
Solution Approach 2:
The patent merges multiple local maximum detections into a unified similarity measure by calculating the cumulative area under the similarity curve. This combining approach integrates information from multiple comparison points to produce a more reliable and accurate match determination.
Data Source
AI summary
Systems and methods are provided for generating unique signatures for digital video files to locate video sequences within a video file comprising calculating a frame signature for each frame of a first video; and for a second video: calculating a frame signature for each frame of the second video for corresponding first video frame signatures, calculating a frame distance between each of the corresponding video frame signatures, determining video signature similarity between the videos, and searching within a video signature similarity curve to determine a maximum corresponding to the first video within the second video. The method further applies area augmentation to the video signature similarity curve to determine a maximum from among a plurality of maxima corresponding to the first video file within the second video file.