Dual-Stage Image Retrieval for Semantic and Pixel Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image retrieval methods are ineffective and labor-intensive, making it difficult to retrieve image materials that meet user requirements in terms of both semantic and pixel-level accuracy.
Innovation Solution
An image retrieval method that utilizes two-dimensional retrieval reference information, including semantic and pixel information, to efficiently match target sample images by first extracting semantic and pixel features from a reference image and then using semantic and pixel information libraries for staged retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image retrieval methods are used, then the retrieval process is simple, but the retrieval accuracy and effectiveness are poor
Solution Approach 1:
The patent divides the image retrieval process into two distinct stages: semantic information retrieval and pixel information retrieval. The semantic retrieval stage first filters candidate images based on semantic features extracted from the reference image, and then the pixel retrieval stage further refines the results by comparing pixel-level features. This segmentation allows each stage to focus on specific aspects of image similarity, improving overall retrieval accuracy without requiring a single complex system to handle all aspects simultaneously.
Solution Approach 2:
The patent extends the retrieval process from a single-dimension approach to a two-dimensional approach by incorporating both semantic information (content meaning) and pixel information (visual appearance) as separate retrieval dimensions. This dimensional expansion allows the system to evaluate images from multiple perspectives, significantly improving retrieval precision by considering both what the image represents and how it visually appears.
2Measurement precision
If manual image retrieval is performed to ensure quality, then retrieval accuracy improves, but labor cost and time consumption increase
Solution Approach 1:
The patent replaces manual mechanical image retrieval operations with an automated computer-based system that performs semantic feature extraction, pixel feature extraction, and similarity comparison operations. This substitution eliminates the need for manual inspection and selection of images, dramatically reducing time consumption and labor costs while maintaining high retrieval accuracy through systematic algorithmic processing of both semantic and visual features.
Solution Approach 2:
The system enables automated self-service image retrieval by using algorithms to automatically extract features, compare images, and select results without human intervention. The dual-stage retrieval process autonomously handles the entire workflow from reference image input to result output, allowing users to obtain high-quality retrieval results without investing time in manual search and evaluation processes.
3Productivity
If single-dimension retrieval is used, then the retrieval process is fast, but the retrieved images cannot meet user requirements for both semantics and visuals
Solution Approach 1:
The patent segments the retrieval task into two efficient sequential stages: semantic retrieval that quickly filters candidates based on content meaning, and pixel retrieval that efficiently refines results based on visual similarity. This segmentation allows each stage to operate optimally within its specific domain, maintaining high processing speed while comprehensively evaluating both semantic and visual aspects of images.
Solution Approach 2:
The system achieves comprehensive retrieval quality by operating in two dimensions simultaneously: semantic dimension for content understanding and pixel dimension for visual appearance. This two-dimensional approach ensures that retrieved images satisfy both semantic relevance and visual similarity requirements, overcoming the limitations of single-dimension retrieval that can only optimize for speed or quality in one aspect.
Data Source
AI summary
The present disclosure relates to an image retrieval method, an electronic device, and non-transitory computer-readable storage medium. The method includes: obtaining retrieval reference information, where the retrieval reference information includes first semantic information and first pixel information; based on the first semantic information, retrieving from a preset semantic information library and obtaining multiple candidate semantic information that match the first semantic information; based on candidate sample images to which the multiple candidate semantic information respectively belong, retrieving from a preset pixel information library and obtaining candidate pixel information corresponding to each of the multiple candidate sample images; according to a similarity between the candidate pixel information corresponding to each of the multiple candidate sample images and the first pixel information, selecting a target sample image from the multiple candidate sample images, and taking the target sample image as a retrieval result image corresponding to the retrieval reference information.