Image Retrieval Accuracy via Dual-Space Vector Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image retrieval systems often retrieve images with different concepts due to solely relying on image data similarity calculations, leading to noise in retrieval results and decreased accuracy.
Innovation Solution
An image retrieval system and method that utilize a database with image feature value data and tag data, where neural networks are employed to generate feature vectors and tag vectors, and calculate similarities to correct and enhance retrieval accuracy by linking query tags to database images based on similarity, using cosine similarity calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image retrieval is performed using only image data similarity calculation, then retrieval speed is maintained, but retrieval accuracy deteriorates due to inclusion of images with different concepts
Solution Approach 1:
The patent introduces a tag space dimension complementing the traditional image feature space. By representing both images and tags as vectors in their respective spaces and calculating similarities in both dimensions, the system retrieves images that are similar both visually and conceptually, thereby improving retrieval accuracy without excessive complexity increase
Solution Approach 2:
The patent segments the retrieval process into two independent similarity calculations: image-space similarity (visual features) and tag-space similarity (conceptual features). These segmented calculations are then combined to produce the final retrieval result, allowing each dimension to be optimized independently while improving overall accuracy
2Measurement precision
If multiple types of data are processed to improve retrieval accuracy, then retrieval accuracy improves, but processing time increases
Solution Approach 1:
The patent performs preliminary processing by pre-computing and storing tag vectors for all database images along with their image feature vectors. When a retrieval query arrives, the system only needs to perform vector similarity calculations using pre-computed data, significantly reducing processing time compared to generating tags on-demand
Solution Approach 2:
The patent replaces complex semantic understanding and conceptual analysis with vector space model and cosine similarity calculations. By representing tags and images as mathematical vectors and using efficient similarity metrics, the system achieves accurate conceptual matching without time-consuming manual or complex computational analysis
Data Source
AI summary
An image retrieval system with high retrieval accuracy is provided. The image retrieval system includes a database and a processing portion. The database has a function of storing a plurality of pieces of database image data, and a database tag is linked to each of the plurality of pieces of database image data. The processing portion has a function of obtaining database image feature value data representing a feature value of the database image data for each piece of the database image data. The processing portion has a function of obtaining query image feature value data representing a feature value of the query image data. The processing portion has a function of calculating first similarity of the database image data to the query image data for each piece of the database image data. The processing portion has a function of obtaining a query tag linked to the query image data using some of the database tags.