Image Search Query Correction via Relevance Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search technologies fail to accurately reflect a user's subjective search intention, as they do not generate new queries based on the relevance between multiple images hit by the search, leading to inefficiencies in high-speed search and classification of large-scale image data.
Innovation Solution
An image search device with a search condition input unit, query generation unit, image search unit, relevance estimation unit, and query correction unit that generates and refines queries based on user input and image relevance, allowing for accurate reflection of search intention through iterative attribute refinement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If image search is performed using traditional search methods, then search speed is improved, but the accuracy of reflecting user search intention deteriorates
Solution Approach 1:
The system implements feedback by estimating the relevance between a plurality of images and using this relevance information to correct and refine search queries. The relevance estimation unit analyzes the relationship between retrieved images, and the query correction unit generates improved queries based on this analysis, creating a closed-loop system that continuously improves search accuracy while maintaining speed.
Solution Approach 2:
The system performs preliminary actions by pre-processing image data to extract features and metadata before actual search operations. It also pre-calculates relevance relationships between images to enable faster query corrections during the search process, reducing the computational burden during real-time search operations.
2Measurement precision
If manual addition of text information is performed, then search accuracy is improved, but time consumption increases
Solution Approach 1:
The system performs self-service by automatically adding text information (metadata) to images through image recognition technology. The automatic metadata generation eliminates the need for manual annotation while maintaining high search accuracy, as the system autonomously extracts and attaches relevant descriptive information to images based on their visual content.
Solution Approach 2:
The system replaces the mechanical manual annotation process with automated image recognition technology. Instead of requiring human operators to manually add text information, the system uses computer vision and machine learning algorithms to automatically extract and generate metadata, significantly reducing time consumption while maintaining or improving search accuracy.
3Quantity of substance
If a large number of images are searched, then search completeness is improved, but the difficulty of classifying and filtering results increases
Solution Approach 1:
The system segments the large set of search results by estimating relevance relationships between images and using this information to divide the results into meaningful groups or categories. The relevance estimation unit analyzes relationships between images, enabling the system to segment and organize large result sets into manageable classifications that are easier to navigate and filter.
Solution Approach 2:
The system changes parameters by dynamically adjusting search criteria and relevance thresholds based on the analyzed relationships between images. The query correction unit modifies search parameters according to relevance estimation results, enabling effective classification and filtering of large result sets by adapting search parameters to the specific characteristics of the retrieved images.
Data Source
AI summary
The invention is directed to an image search device including a search condition input unit that receives a search condition for searching for an image, a query generation unit that generates a first query based on the search condition, an image search unit that searches for an image in a database based on the first query, a relevance estimation unit that estimates relevance between a plurality of images selected by a predetermined operation among images hit by a search, a query correction unit that generates a second query based on the relevance between the plurality of images, and a function of displaying the second query generated by the query correction unit on an interface.