Compact Visual Search Descriptor via Graph Spectral Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile visual search systems face high computational costs and communication expenses due to the large size of visual descriptors, such as SIFT and SURF feature points, which hinder real-time operations and efficient data transmission.
Innovation Solution
A compression scheme using graph spectral embedding models reduces the dimensionality of feature points while preserving pair-wise nearest neighbor relationships, employing a linear embedding operator and affinity matrix to generate a second set of feature points that can be transmitted efficiently over wireless networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-dimensional visual descriptors (SIFT/SURF) are used to ensure accurate image matching, then matching precision is improved, but communication cost and data transmission size increase
Solution Approach 1:
The patent extracts only the most essential features from high-dimensional descriptors by identifying and retaining extremal points (local maxima/minima) in the feature space. This extraction process removes redundant dimensions while preserving the core discriminative information needed for accurate matching, directly resolving the contradiction between descriptor size and matching precision.
Solution Approach 2:
The patent creates a compact copy of the original high-dimensional descriptors by mapping them to a lower-dimensional space using learned transformation matrices. This copy retains the essential matching information while occupying significantly less space, enabling efficient transmission without sacrificing matching accuracy.
2Measurement precision
If high-dimensional visual descriptors are transmitted over wireless networks, then matching accuracy is maintained, but communication cost increases
Solution Approach 1:
The patent extracts only the essential discriminative features from complete descriptors before transmission. By removing redundant dimensional information that is not critical for accurate matching, the communication payload is reduced while maintaining query accuracy, thereby lowering wireless transmission energy costs.
Solution Approach 2:
The patent transforms descriptors from high-dimensional parameter spaces to low-dimensional parameter spaces through learned linear mappings. This parameter transformation reduces the amount of data that needs to be transmitted over wireless networks while preserving the essential information needed for accurate visual search queries.
3Reliability
If complete feature points are processed, then matching performance is preserved, but computational complexity increases
Solution Approach 1:
The patent extracts a compact representation from complete feature points by identifying extremal points in the feature space and retaining only those critical for matching. This extraction reduces the number of features that need to be processed while maintaining matching performance, thereby reducing computational complexity on mobile devices.
Solution Approach 2:
The patent performs preliminary dimensionality reduction and feature extraction before the actual matching process. By pre-processing descriptors to remove redundant information and transform them into compact forms, the subsequent matching operations become computationally less intensive while preserving matching performance.
Data Source
AI summary
A method and an apparatus for processing an image generates, using a linear embedding operator, a second set of feature points from a first set of feature points of the image, so that a pair-wise nearest neighbor (NN) topological relationship among the first set of feature points is preserved in the second set of feature points. The linear embedding operator is determined by an affinity model comprising a first affinity parameter, a second affinity parameter, and an affinity matrix, wherein a sparsity of the affinity matrix is controlled by the first affinity parameter and the second affinity parameter.