Image Search Result Grouping by Label Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search engines often provide inaccurate results when users attempt to identify objects in images, as they rely on text queries that may not accurately represent the image content, and user-generated content submissions with varying annotations can lead to inconsistent search results.
Innovation Solution
A computer-implemented method and system that receives image queries, identifies matching images from user submissions, determines label similarity, groups results based on label similarity, assigns image match scores, and adjusts scores based on geographic information, ensuring that only the best result from each user is presented, while grouping and ranking results for accurate and relevant object identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multiple content submissions with varying annotations are used, then the quantity of search results increases, but the accuracy and consistency of results deteriorates
Solution Approach 1:
The patent segments search results into groups based on label similarity, where each group contains submissions with similar annotations. This segmentation allows the system to maintain multiple results (satisfying quantity requirement) while organizing them into coherent groups with consistent labels (maintaining accuracy and consistency within each group).
Solution Approach 2:
The patent changes the parameter of label similarity by computing similarity metrics between annotations and using these to determine grouping. By adjusting the similarity threshold and grouping parameters, the system can control both the quantity of results presented and the consistency of labels within groups, resolving the contradiction between quantity and accuracy.
2Loss of information
If all content submissions are presented as results, then the completeness of information increases, but the complexity of result presentation increases
Solution Approach 1:
The patent segments the complete set of search results into labeled groups based on annotation similarity. This segmentation preserves all information (completeness) by including all submissions in the search, while organizing them into manageable groups with consistent labels that simplify presentation and user comprehension.
Solution Approach 2:
The patent merges multiple content submissions with similar labels into single groups, presenting them as unified result categories. This merging reduces presentation complexity by consolidating similar items while maintaining completeness through the inclusion of all original submissions within their respective groups.
3Ease of operation
If text queries are used to search images, then the ease of operation increases, but the accuracy of identifying image content deteriorates
Solution Approach 1:
The patent introduces user-generated labels as an intermediary between text queries and image content identification. Instead of directly matching text queries to image features, the system uses community-provided labels as a mediating layer that bridges the gap between simple text input and accurate image identification, maintaining ease of operation while improving accuracy.
Solution Approach 2:
The patent implements feedback mechanisms where user annotations and labels are continuously refined based on similarity analysis and grouping. This feedback loop allows the system to learn from community input, improving the accuracy of image content identification over time while maintaining the simplicity of text-based search operations.
Data Source
AI summary
A system and computer-implemented method is provided for organizing multiple user submitted results responsive to an image query. A plurality of content submissions may be received from a variety of submitting users, each content submission including an image and an associated label. An image query may provide an image of an object as a request to identify the object. In response to receiving the image query, one or more results of the plurality of content submissions may be identified. A similarity between the labels for each of the one or more results may be determined and used to group the one or more results. Grouped results may be ranked and sorted for accurate and concise presentation to a querying user.