Image Retrieval Using Composite Global and Content Feature Codes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image retrieval methods based on global features often fail to accurately retrieve images with prominent or core content, such as art paintings, as they prioritize similarity in color and texture over content similarity, leading to low retrieval accuracy and efficiency.
Innovation Solution
An image retrieval method that extracts both global and content features using convolutional neural networks, converts them into binarized feature codes, and combines these codes to create a composite feature code for comparison with alternative codes in a library, enhancing retrieval accuracy by considering both overall and core content similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If global feature-based retrieval is used, then retrieval speed is improved, but retrieval accuracy deteriorates for images with prominent content
Solution Approach 1:
The patent divides the image into global features and content features separately. Global features capture overall characteristics while content features focus on prominent objects. By segmenting the feature extraction process, the system achieves both fast global matching and accurate content-based retrieval.
Solution Approach 2:
The patent combines global feature codes and content feature codes into a composite feature code. This composite representation integrates the advantages of both global features (speed) and content features (accuracy), enabling efficient and precise image retrieval simultaneously.
2Device complexity
If only global features are extracted, then device complexity is reduced, but retrieval accuracy deteriorates
Solution Approach 1:
The feature extraction process is segmented into two independent modules: global feature extraction and content feature extraction. Each module uses convolutional neural networks but processes different aspects of the image, allowing complexity to be divided and managed separately while improving overall accuracy.
Solution Approach 2:
The patent merges global feature codes and content feature codes into a unified composite feature code. This combination integrates multiple feature types without significantly increasing device complexity, as both features are processed through similar neural network architectures.
3Measurement precision
If content features are extracted in addition to global features, then retrieval accuracy is improved, but processing time increases
Solution Approach 1:
The patent transforms continuous feature vectors into discrete binarized codes using hashing algorithms. This parameter transformation reduces the dimensionality and complexity of feature comparison, enabling fast Hamming distance calculations that minimize processing time while maintaining high retrieval accuracy.
Solution Approach 2:
The patent creates compact binary representations (copies) of the complex feature vectors. These binarized feature codes serve as simplified copies that retain essential information for accurate retrieval while enabling extremely fast comparison operations.
Data Source
AI summary
Disclosed are an image retrieval method, an image retrieval apparatus, an image retrieval device and a computer-readable storage medium, wherein the image retrieval method including: extracting global features of an input image and obtaining a global feature code based on the global features; extracting content features of the input image and obtaining a content feature code based on the content features; splicing the global feature code with the content feature code to obtain a composite feature code of the input image; comparing the composite feature code with alternative feature codes in a feature code library to obtain an image retrieval result. By comprehensively considering the global features and the content features of the input image in the image retrieval process, the accuracy and retrieval efficiency of the retrieval results can be improved.