Image Search Using Conspicuous Area Detection for Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image search systems struggle to accurately find specific moving objects, such as people, in large video datasets due to reliance on conspicuous features like red clothing, leading to false positives and missed targets when many individuals wear similar colors, as they do not consider the relationship between image features and other people in the scene.
Innovation Solution
An image search system that learns search priority by analyzing areas in target images and query images, detecting conspicuous areas with high search priority, and using image features from these areas to enhance search accuracy by focusing on low appearance frequency features like shoes and bags.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If image features are extracted from conspicuous areas (e.g., red clothes), then search speed is improved, but search accuracy deteriorates due to false positives from multiple people wearing similar colors
Solution Approach 1:
The patent segments the image into multiple areas (conspicuous areas and non-conspicuous areas) and extracts image features from both segments. By dividing the search into multiple feature extraction zones, the system balances between using conspicuous features for speed and non-conspicuous features for accuracy, resolving the contradiction between search speed and search accuracy.
2Ease of operation
If only conspicuous features are used for search, then ease of operation is improved, but reliability deteriorates due to inability to distinguish target from similar objects
Solution Approach 1:
The patent applies local quality by assigning different weights and priorities to different areas of the image. Conspicuous areas are given higher weight for ease of detection, while non-conspicuous areas are also considered to ensure reliability. This local differentiation allows the system to maintain operational simplicity while improving search reliability through multi-area feature extraction.
3Measurement precision
If image features are extracted from all areas, then search accuracy is improved, but device complexity increases due to multiple analysis requirements
Solution Approach 1:
The patent extracts only the necessary image features from specific areas rather than processing all image data uniformly. By taking out and analyzing only conspicuous and non-conspicuous areas separately, the system achieves high search accuracy while controlling device complexity through selective feature extraction rather than comprehensive image processing.
4Productivity
If conventional similarity-based search is used, then productivity is improved, but measurement precision deteriorates due to false positives in crowded scenes
Solution Approach 1:
The patent performs preliminary analysis to identify both conspicuous and non-conspicuous areas before conducting the similarity-based search. This preliminary action of area identification and feature extraction preparation enables the subsequent search to be both efficient (maintaining productivity) and accurate (improving measurement precision) by having pre-processed feature data from multiple image regions ready for comparison.
Data Source
AI summary
An image search apparatus including: an accumulation unit that accumulates search target images; an area detector that learns search priority by analyzing first information extracted from a plurality of first areas included in a plurality of the search target images, analyzes second information extracted from a plurality of second areas included in a query image, and detects the second areas, the search priority of which is high, as conspicuous areas by using the second information; and an image search unit that searches for the search target images by using an image feature extracted from the conspicuous areas.