Weighted Min-Hash for Cropped Media Near-Duplicate Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional near-duplicate detection and hashing schemes are unreliable when dealing with spatially or temporally cropped media files, as they are sensitive to variations in position and temporal information, leading to incongruous feature descriptors and ineffective detection of near-duplicates.
Innovation Solution
A robust hashing method is applied to media data using a weighted min-hash procedure, generating a histogram based on feature information and creating randomized versions of media files to encode frequency of features, which are then hashed using seeded hash functions, making the method robust to variations in media data dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional hashing schemes are used for near-duplicate detection, then the system can process media files, but the detection reliability deteriorates when media files are spatially or temporally cropped versions of each other
Solution Approach 1:
The patent segments the media data into multiple randomized versions by applying different crop amounts and positions to the media file. This creates multiple variations of the same media content, allowing the histogram to capture feature distributions that are robust to cropping. By segmenting the data this way, the system can identify near-duplicates even when the original files have different crop configurations.
Solution Approach 2:
The patent changes the parameters of the media data by creating randomized versions with different crop amounts and positions. These parameter changes transform the original media file into multiple variations, enabling the histogram-based hashing to capture invariant features that remain consistent across different crop configurations, thereby improving detection reliability despite cropping variations.
2Measurement precision
If position information and temporal information are used in feature descriptors, then the system can characterize media data, but the feature descriptors become incongruous when media files are cropped
Solution Approach 1:
The patent segments the feature descriptors into multiple randomized versions by applying different crop configurations to the media data. This segmentation allows the system to collect feature descriptors from multiple perspectives and crop configurations, then aggregate them into a histogram that represents the overall feature distribution, making the feature representation stable despite variations in position and temporal information.
Solution Approach 2:
The patent creates multiple copies of the media data with different crop amounts and positions. These copies serve as alternative representations of the same media content, allowing the system to gather feature descriptors from multiple copies and aggregate them into a consistent histogram representation, thereby maintaining feature descriptor consistency across cropped versions.
3Reliability
If the histogram is generated from multiple randomized versions of the media file, then the hashing becomes robust to cropping variations, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by generating a limited number of randomized versions (e.g., 10 versions with different crop amounts and positions) rather than exhaustively generating all possible variations. This partial sampling is sufficient to capture the robust features needed for reliable near-duplicate detection while keeping the computational complexity manageable. The histogram aggregates features from these partial samples, achieving robustness without requiring complete enumeration of all possible crop configurations.
Data Source
AI summary
A robust hashing method is applied to media data (e.g., video, image, and/or audio data), producing a hash output that is robust with respect to at least one attribute of the media data. A histogram is generated for the media data and the histogram is hashed using a weighted hashing procedure. The histogram can be derived from a plurality of randomized versions of the media file, each randomized version of the media file altered to a random extent with respect to the attribute. The histogram can also be derived from a plurality of feature descriptors computed for the media data that are coarsely encoded with respect to the attribute. The weighted hashing procedure includes assigning a weight to components of the histogram and applying a plurality of hash functions to a number of versions of each component, the number of versions based on the assigned weight.