LSH Video Indexing Sub-sampling for Lookup Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing indexing systems face inefficiencies and accuracy issues when handling large video databases, particularly in online video sharing sites, due to the return of very large lists of results during index lookups, which can overwhelm processing power, memory, and network bandwidth, leading to decreased matching accuracy when blacklisting common queries is used.
Innovation Solution
A system that indexes reference videos using Locality Sensitive Hashing (LSH) and sub-sampling techniques, where reference identifiers are stored in bins and sub-sampled based on a predefined threshold, using mapping functions to manage the number of results and maintain accuracy, allowing for efficient lookup and retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional indexing returns all matching reference videos, then completeness of results is improved, but processing efficiency and system performance deteriorate due to overwhelming data flow
Solution Approach 1:
The patent applies partial action by returning only a subset of matching reference videos instead of all matches. The system uses sub-sampling to select a representative portion of results, which maintains adequate matching accuracy while significantly reducing processing load and improving system performance.
Solution Approach 2:
The patent changes the parameter of result quantity from 'all matches' to 'sub-sampled matches'. By introducing a sub-sampling factor that controls the ratio of returned results, the system optimizes the balance between result completeness and processing efficiency, transforming an overwhelming data flow into a manageable subset.
2Productivity
If blacklisting common input queries is used to reduce data flow, then processing efficiency is improved, but matching accuracy deteriorates due to complete loss of information
Solution Approach 1:
Instead of completely discarding common queries through blacklisting, the patent uses partial action by sub-sampling the results. This returns a representative subset of matches for common queries, maintaining matching accuracy while still reducing processing load compared to returning all matches.
Solution Approach 2:
The patent extracts a sub-sampled subset of results from the complete match set. Rather than removing entire query categories (blacklisting), it extracts a manageable portion of results that preserves the essential matching information while reducing data flow.
3Productivity
If sub-sampling is applied to limit results, then processing efficiency is improved, but result completeness deteriorates
Solution Approach 1:
The patent introduces and optimizes the sub-sampling factor parameter to control the balance between efficiency and completeness. By adjusting this parameter, the system can adapt to different scenarios, ensuring that enough results are returned to maintain matching accuracy while keeping processing efficient.
Solution Approach 2:
The system applies partial action by returning a subset of results that is sufficient for practical purposes. The sub-sampled results, while not exhaustive, contain enough information to achieve adequate matching accuracy, making the loss of completeness acceptable in exchange for significant efficiency gains.
Data Source
AI summary
A set of reference videos is indexed to a reference index in order to facilitate matching of video content. An indexing module receives a set of reference fingerprints representing a set of reference videos and identifies keys contained in the reference fingerprints. Reference identifiers for the reference videos are stored in bins of the reference index associated with the identified keys. The bins in the reference index are sub-sampled to limit the number of reference identifiers stored in a given bin.