Progressive Visual Search Query Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mobile visual search systems face high computational costs and communication expenses due to the large size of visual descriptors, which hinder real-time operations and compromise matching accuracy when reducing bit rate.
Innovation Solution
The implementation of progressive queries in visual search applications, where a sequence of queries based on feature points and their locations is sent over the network, allowing for partial matching without transmitting the complete set of feature points, using methods like dimension reduction, hashing, and topological encoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complete visual descriptors (feature points) are transmitted for visual search, then matching accuracy is improved, but communication cost and data size increase significantly
Solution Approach 1:
The complete set of feature points is divided into multiple subsets or groups. Instead of transmitting all feature points at once, the system segments them into manageable portions that can be transmitted sequentially or selectively, reducing the data size of each transmission while maintaining overall matching accuracy through cumulative processing.
Solution Approach 2:
The system extracts only the most discriminative or representative feature points from the complete set rather than transmitting all feature points. This selective extraction maintains matching accuracy by focusing on the most informative features while significantly reducing the quantity of data transmitted.
2Measurement precision
If complete visual descriptors are transmitted, then matching accuracy is improved, but computational cost and processing time increase
Solution Approach 1:
The computational process is segmented into multiple stages corresponding to different subsets of feature points. The system processes feature points in groups rather than all at once, allowing intermediate results to be evaluated and potentially terminated early if sufficient matches are found, thereby reducing overall computational cost while maintaining accuracy.
Solution Approach 2:
The system performs partial processing by evaluating only a subset of feature points when sufficient matching information is obtained. Instead of exhaustively processing all feature points, the method accepts partial action (processing only necessary subsets) to achieve satisfactory matching accuracy with reduced computational burden.
3Measurement precision
If complete feature points are sent for visual search, then matching accuracy is improved, but real-time operation capability deteriorates
Solution Approach 1:
The system performs preliminary evaluation with a first subset of feature points before committing to complete transmission. By conducting initial matching operations with partial data and evaluating intermediate results, the system can determine whether full feature point transmission is necessary, enabling real-time operation through early termination when sufficient matches are found.
Solution Approach 2:
The method implements partial action by transmitting and processing only the necessary subset of feature points for achieving satisfactory matching accuracy in real-time. Instead of always transmitting complete feature sets, the system adapts the transmission volume based on real-time requirements, thereby maintaining productivity while preserving adequate matching performance.
Data Source
Figure 1(a)~1(b)
Figure 2~3
Figure 4(a)~4(b)
AI summary
A system and a method for progressive queries, which are a sequence of queries for visual search of an image, sent from a sender to a receiver of a communication network, are disclosed. Queries based on a set of feature points and their associated locations of an image may be selected in sequence. A sender sends a first query to a receiver, wherein the first query comprises a number of feature points which is a subset of the set of feature points of the image invariant to rotation and scaling in creating the image. The sender receives a feedback from the receiver, wherein the feedback is generated by the receiver based on searching an image repository using information from the first query. The sender then decides whether there is a match of the image in the image repository based on the received feedback.