High-Dimensional Image Feature Matching via Layered Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-dimensional image feature matching is challenging due to the curse of dimensionality, making it difficult to efficiently match high-dimensional image features in databases, especially when using exhaustive search or binary code learning methods.
Innovation Solution
The method involves extracting high-dimensional image features, dividing them into low-dimensional features, determining nearest clustering centers using layered clustering calculations, and calculating similarities to retrieve matching high-dimensional features through a k-means tree and TF-IDF scoring, with quick select and insertion sort algorithms for efficient retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exhaustive search is used to match high-dimensional features, then matching accuracy is improved, but computational complexity and time consumption increase significantly
Solution Approach 1:
The patent divides high-dimensional feature vectors into multiple low-dimensional sub-features (e.g., splitting an 8192-dimensional feature into multiple 64-dimensional sub-features). This segmentation allows the system to perform matching on reduced-dimensional data while maintaining matching accuracy, thereby reducing computational complexity and time consumption without sacrificing precision.
2Productivity
If binary code learning is used to reduce dimensionality, then computational efficiency is improved, but matching accuracy deteriorates due to information loss
Solution Approach 1:
Instead of converting high-dimensional features to binary codes (which loses information), the patent transforms the problem by working with multiple low-dimensional sub-features simultaneously. This dimensional transformation preserves the original information while enabling efficient computation through operations on smaller vectors, avoiding the accuracy loss inherent in binary code conversion.
3Loss of information
If high-dimensional features are directly matched, then feature representation completeness is improved, but retrieval time increases due to curse of dimensionality
Solution Approach 1:
The patent segments high-dimensional feature vectors into multiple low-dimensional sub-features, which can be processed independently and in parallel. This segmentation maintains the completeness of feature representation (all sub-features together represent the original high-dimensional feature) while dramatically reducing retrieval time by enabling efficient low-dimensional operations and parallel processing.
Data Source
AI summary
A high-dimensional image feature matching method and device relating to the field of image retrieval. The method includes extracting a high-dimensional image feature of an image to be retrieved; dividing the high-dimensional image feature of the image to be retrieved into a plurality of low-dimensional image features; comparing each of the low-dimensional image features of the image to be retrieved with clustering centers at each layer of the low-dimensional image features of the images in a database; and determining a similarity the low-dimensional image feature between the image to be retrieved and each of some images in the database according to a comparison result, so that at least one feature matching the high-dimensional image feature of the image to be retrieved is retrieved in the database.