Two-Stage Image Retrieval Using Deep Features and Keypoint Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image retrieval methods face challenges in achieving high retrieval efficiency and precision, particularly in identifying similar images from large datasets.
Innovation Solution
A two-layer algorithmic approach combining deep learning-based and non-deep learning-based feature extraction methods, utilizing a semi-supervised learning model to enhance depth feature quality and manual feature verification for precise image matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning-based feature extraction is used, then retrieval precision is improved, but computational load and processing time increase
Solution Approach 1:
The patent divides the feature extraction process into two distinct segments: deep learning-based feature extraction for initial candidate selection, and traditional manual feature extraction for final verification. This segmentation allows each method to be applied where it is most effective, reducing overall processing time while maintaining high retrieval precision.
Solution Approach 2:
The patent performs deep learning-based feature extraction as a preliminary action to quickly identify candidate images from the full dataset. This preliminary filtering reduces the number of images that require more computationally intensive traditional feature extraction and matching, thereby reducing total processing time.
2Productivity
If traditional manual feature extraction is used, then processing speed is improved, but retrieval precision deteriorates
Solution Approach 1:
The patent segments the retrieval process into two stages: deep learning-based feature extraction for candidate selection, and traditional manual feature extraction for verification. This allows the system to leverage the speed of traditional methods while using deep learning to ensure high precision in the initial selection phase.
Solution Approach 2:
The patent uses deep learning-based features as an intermediary to bridge the gap between speed and precision. The deep learning features serve as a preliminary filter that identifies promising candidates, which are then verified using traditional manual features, combining the advantages of both approaches.
3Measurement precision
If deep learning models are trained with large datasets, then feature quality is improved, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-training the deep learning model on large datasets beforehand. Once trained, the model can be deployed for inference without requiring the same computational resources during actual retrieval operations, separating the training time cost from the operational time cost.
Data Source
AI summary
An image retrieval method is provided. The image retrieval method includes: performing feature extraction on an image according to a deep learning-based feature extraction manner to obtain a first feature; determining, based on the first feature, at least one candidate image; performing feature extraction on the image and each of the at least one candidate image according to a non-deep learning-based extraction manner to obtain a second feature; sequentially performing key point matching processing on the image with each of the at least one candidate image based on the extracted second feature; determining a first image, from among the at least one candidate image, as an image retrieval result, based on a quantity of matched key points between the first image and the image being greater than a quantity threshold; and transmitting the image retrieval result.