Image Search Feature Vector Clustering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional image searching methods face challenges in accuracy and efficiency due to the difficulty in describing images with keywords, varying user assignments of keywords, and the inefficiency of feature vector-based approaches with large image datasets.

Innovation Solution

A system and method that determine a target feature vector for a target image and feature vectors for multiple images, creating a search region in an N-dimensional feature space based on a preset similarity range to identify similar images by selecting candidate images within this region and ranking them by similarity degree.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If feature vector-based image searching is used, then image searching accuracy is improved, but searching efficiency deteriorates when the number of images is large

Engineering Contradiction:
Improveimage searching accuracyVSAvoidsearching efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the image database into multiple clusters based on feature vectors, creating a hierarchical search structure. Instead of comparing the target image with all images in the database, the system first identifies relevant clusters and then performs detailed comparisons only within those clusters. This segmentation approach maintains high search accuracy while dramatically improving efficiency by reducing the number of comparisons needed in large datasets.

Inventive Principle:
Principle #1Segmentation

2Productivity

If textual meta-information is used for image searching, then searching speed is improved, but searching accuracy deteriorates

Engineering Contradiction:
Improvesearching speedVSAvoidsearching accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary extraction of feature vectors from images and organizes them into clusters before the actual search operation. This preliminary action creates a structured framework that enables fast retrieval. During the search phase, the system uses this pre-organized structure to quickly identify relevant image groups, achieving both high speed and high accuracy by combining efficient preprocessing with targeted comparison.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3776261B1Systems and methods for image searching
Publication Date: 2023.07.12 ZHEJIANG DAHUA TECH CO LTD
  • EP3776261B1 patent drawingFigure 1
  • EP3776261B1 patent drawingFigure 2
  • EP3776261B1 patent drawingFigure 3

AI summary

A system and a method for image searching. The method may include obtaining a target image and a plurality of images, wherein at least one similar image to the target image is to be searched from the plurality of images. The method may include determining a target feature vector of the target image and a feature vector of each of the plurality of images, each of the target feature vector and the feature vectors being an N-dimensional feature vector in an N-dimensional feature space. The method may include determining a search region in the N-dimensional feature space. The method may include determining one or more candidate images, the feature vector of each candidate image being within the search region. The method may further include determining the at least one similar image to the target image among the one or more candidate images.