Region-Aware Image Retrieval Using Text-Semantic Texture Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image retrieval methods suffer from low accuracy due to the difficulty in distinguishing and locating objects of interest within images, as they rely on global semantic features that are often interfered by irrelevant regions, leading to retrieval results that do not align with user intentions.
Innovation Solution
A method that extracts a first text feature from user input, determines a first semantic feature of a specific region in an image, generates a second semantic feature based on the text and image features, and uses a combined visual query feature group comprising both semantic and texture features to accurately locate target images in a candidate set.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If global semantic features are used for image retrieval, then the retrieval process is simple and fast, but the retrieval accuracy deteriorates due to interference from irrelevant regions
Solution Approach 1:
The patent segments the image into multiple regions and extracts semantic features for each region separately, then combines them with region-level texture features. This segmentation allows the system to focus on relevant regions while maintaining computational efficiency, resolving the contradiction between simple processing and accurate retrieval.
Solution Approach 2:
The patent applies local quality by extracting and utilizing texture features specific to each region rather than using a single global texture representation. This allows different regions to contribute their specific texture characteristics to the retrieval process, improving accuracy without significantly increasing computational complexity.
2Measurement precision
If region-specific semantic features are extracted, then retrieval accuracy improves by focusing on relevant regions, but device complexity increases due to additional processing steps
Solution Approach 1:
The patent merges region-level semantic features with region-level texture features to create a comprehensive feature representation. By combining these feature types at the region level, the system achieves improved accuracy without requiring overly complex processing, as the merging operation integrates multiple information sources efficiently.
3Measurement precision
If multiple feature types (semantic and texture) are combined for each region, then retrieval accuracy improves, but computational time increases
Solution Approach 1:
The patent segments the image into regions and processes semantic and texture features independently at each region level, then combines the results. This segmented approach allows for efficient parallel processing and reduces the computational burden compared to processing all features simultaneously across the entire image.
Solution Approach 2:
The patent extracts and processes only the necessary semantic and texture features for each region rather than extracting all possible features. This partial action approach optimizes computational time by focusing resources on the most relevant features while maintaining high retrieval accuracy.
Data Source
AI summary
A method executed by an electronic device includes, extracting a first text feature from a first text input by a user, extracting, from a first image input by the user, a first semantic feature of a first region related to the first text; generating a second semantic feature based on the first text feature and the first semantic feature; and obtaining an image from a candidate image set based on the second semantic feature and a first texture feature of the first region.