Feature Descriptor Compression for Scalable Image Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing volume of images and feature descriptors in databases for image matching and content retrieval poses a storage scalability challenge, as existing methods require substantial storage space and can lead to performance degradation and information loss due to compression.
Innovation Solution
The system compresses feature descriptors by clustering them using algorithms like Hierarchical K-means, approximating each cluster with a center, and encoding these centers with fewer bits, allowing for efficient storage and retrieval while using a machine learning technique to compensate for information loss during compression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If feature descriptors are compressed using clustering algorithms, then storage requirements are reduced, but information loss and performance degradation occur
Solution Approach 1:
The feature descriptor is segmented into two parts: a compressed cluster center representation for storage and a residual component for accuracy recovery. During compression, only the cluster center is stored with reduced precision, while the residual (difference between original and cluster center) is preserved and added back during retrieval to reconstruct the full-precision feature descriptor, thus eliminating information loss despite compression.
Solution Approach 2:
The precision parameter of feature descriptor storage is dynamically adjusted by storing at two different precision levels: compressed precision for the cluster center and full precision for the residual. This allows the system to achieve compression while maintaining accuracy by combining both representations during query processing.
2Adaptability or versatility
If more images are stored in the database, then image matching coverage increases, but storage costs and processing time increase
Solution Approach 1:
The database is segmented into two storage components: a compressed feature descriptor database storing only cluster centers for rapid access, and a residual database storing correction information. During querying, the system first performs fast comparison using compressed descriptors, then applies residuals only to top-candidate matches, significantly reducing processing time while maintaining comprehensive image matching coverage.
Solution Approach 2:
The system performs preliminary compression of feature descriptors to cluster centers before storage, creating a pre-processed index that enables fast initial filtering. This preliminary action reduces the dimensionality and size of stored data, allowing the database to scale with more images while keeping query processing time manageable through subsequent residual correction only for promising candidates.
Data Source
AI summary
Various embodiments may increase scalability of image representations stored in a database for use in image matching and retrieval. For example, a system providing image matching can obtain images of a number of inventory items, extract features from each image using a feature extraction algorithm, and transform the same into their feature descriptor representations. These feature descriptor representations can be subsequently stored and used to compare against query images submitted by users. Though the size of each feature descriptor representation isn't particularly large, the total number of these descriptors requires a substantial amount of storage space. Accordingly, feature descriptor representations are compressed to minimize storage and, in one example, machine learning can be used to compensate for information lost as a result of the compression.