Hybrid Visual Search Scoring with Textual Differentiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Visual search technologies often struggle to differentiate between visually similar images, such as wigs and human hair, leading to inaccurate search results when user interaction data is lacking, and fail to effectively combine visual and textual similarities to improve search relevance.
Innovation Solution
The integration of a visual similarity score and a textual similarity score, where the textual similarity is determined based on associated text, such as titles or descriptions, with a weighted combination of both scores to refine search results, allowing user-configurable weighting to adjust the importance of visual versus textual similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual search is used to identify similar images, then visual similarity is improved, but accuracy in differentiating visually similar items deteriorates
Solution Approach 1:
The patent combines visual similarity scoring with textual similarity scoring to create a hybrid search approach. The visual search component identifies images with similar visual characteristics, while the textual search component analyzes associated text descriptions. By merging these two independent scoring systems, the patent resolves the contradiction by maintaining visual similarity matching while adding textual differentiation capability to improve overall search accuracy for visually similar items.
Solution Approach 2:
The patent introduces textual information as an intermediary element that mediates between visual similarity and search accuracy. When visual alone is insufficient to differentiate items (such as wigs versus real hair), the textual descriptions serve as a mediating factor to provide additional discriminative information, thereby improving reliability without compromising visual similarity matching.
2Adaptability or versatility
If only visual similarity is used for search results, then visual relevance is improved, but ability to differentiate items deteriorates
Solution Approach 1:
The patent merges visual similarity assessment with textual similarity assessment to simultaneously achieve visual relevance and item differentiation. The visual component ensures results are visually relevant to the query, while the textual component provides precise differentiation between items. This combination allows the system to be versatile in matching visually similar items while maintaining precision in distinguishing between them through text.
3Adaptability or versatility
If user interaction data is used to improve search, then search personalization is improved, but reliance on data availability deteriorates
Solution Approach 1:
The patent introduces textual information as an intermediary that reduces dependency on user interaction data. The textual similarity component provides a data-independent mechanism for improving search accuracy, allowing the system to function effectively even when user interaction data is unavailable. This intermediary approach maintains reliability while still enabling personalized search through available text metadata.
Data Source
AI summary
Embodiments of the technology described herein, provide improved visual search results by combining a visual similarity and a textual similarity between images. In an embodiment, the visual similarity is quantified as a visual similarity score and the textual similarity is quantified as a textual similarity score. The textual similarity is determined based on text, such as a title, associated with the image. The overall similarity of two images is quantified as a weighted combination of the textual similarity score and the visual similarity score. In an embodiment, the weighting between the textual similarity score and the visual similarity score is user configurable through a control on the search interface. In one embodiment, the aggregate similarity score is the sum of a weighted visual similarity score and a weighted textual similarity score.