Asymmetric Binary Embedding for Image Retrieval Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing binary embedding techniques for image comparison suffer from information loss when compressing query objects, despite reducing computational and memory costs by transforming images into binary spaces.
Innovation Solution
An asymmetric embedding approach is employed, where query objects are represented in a real-valued space using a real-valued embedding function, while database objects are compressed using a quantized embedding function, allowing for efficient comparison without losing query object information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If binary embedding is used to compress database objects, then memory cost and computational cost are reduced, but information loss occurs when query objects are compressed
Solution Approach 1:
The patent applies asymmetric embedding where database objects are compressed into binary space using hash functions, while query objects remain in real-valued space. This asymmetric treatment allows the system to benefit from compressed storage for the large database while maintaining full precision for queries, eliminating information loss in the query path while still achieving memory efficiency for database storage.
2Productivity
If binary embedding is used for both query and database objects, then computational efficiency improves, but accuracy deteriorates due to information loss
Solution Approach 1:
The patent applies local quality by treating database objects and query objects differently based on their roles. Database objects undergo binary compression suitable for efficient storage and batch processing, while query objects maintain real-valued precision for accurate similarity computation. This localized differentiation optimizes each component for its specific function.
3Measurement precision
If real-valued representations are used for all objects, then retrieval accuracy is maintained, but memory cost and response time worsen
Solution Approach 1:
The patent segments the system into two distinct parts: a compressed binary database portion for efficient storage and retrieval, and a real-valued query portion for accurate computation. This segmentation allows the large database to be stored efficiently in binary form while queries are processed in real-valued space, balancing memory efficiency with computational accuracy.
Data Source
AI summary
A system and method for comparing a query object and one or more of a set of database objects are provided. The method includes providing quantized representations of database objects. The database objects have each been transformed with a quantized embedding function which is the composition of a real-valued embedding function and a quantization function. The query object is transformed to a representation of the query object in a real-valued embedding space using the real-valued embedding function. Query-dependent estimated distance values are computed for the query object, based on the transformed query object and stored. A comparison (e.g., distance or similarity) measure between the query object and each of the quantized database object representations is computed based on the stored query-dependent estimated distance values. Data is output based on the comparison computation.