Image Relevance Model Using Visual Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search systems often incorrectly identify images as relevant to queries due to reliance on textual content associated with images, leading to non-relevant search results.
Innovation Solution
An image relevance model is trained using a vector of weights corresponding to content feature values, selecting training images that are relevant to specific queries and others, to generate accurate relevance scores for images based on their content features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If search results are determined based solely on textual content associated with images, then the search system can quickly identify and rank images, but the accuracy of relevance determination deteriorates leading to improper identification of relevant images
Solution Approach 1:
The patent segments the relevance determination process into two independent components: textual content analysis and visual content analysis. The visual content is analyzed separately using image processing techniques to extract visual features, which are then combined with textual features to determine overall relevance. This segmentation allows each component to be optimized independently, maintaining search speed while improving accuracy.
Solution Approach 2:
The patent introduces visual content analysis as an intermediary mechanism between the query and the final relevance determination. Instead of relying solely on textual metadata, the system uses image processing to extract visual features that serve as an intermediate representation, which is then integrated with textual information to produce more accurate relevance scores.
2Productivity
If the search system includes images based on associated textual content, then more images can be retrieved efficiently, but the quality and diversity of search results deteriorates due to inclusion of non-relevant images
Solution Approach 1:
The patent divides the image evaluation process into separate textual and visual analysis streams. By segmenting the assessment into independent visual feature extraction and textual metadata analysis, the system can efficiently process images while maintaining high quality standards through multi-dimensional evaluation.
Solution Approach 2:
The patent changes the parameters used for image evaluation from solely textual metadata to include visual content parameters. By introducing visual feature parameters (such as color histograms, texture features, and structural characteristics) alongside textual parameters, the system achieves more reliable quality assessment without sacrificing retrieval efficiency.
Data Source
AI summary
Methods, systems, and apparatus, including computer program products, for identifying images relevant to a query are disclosed. An image search subsystem selects images to reference in image search results that are responsive to a query based on an image relevance model that is trained for the query. An independent image relevance model is trained for each unique query that is identified by the image search subsystem. The image relevance models can be applied to images to order image search results obtained for the query. Each relevance model is trained based on content feature values of images that are identified as being relevant to the query (e.g., frequently selected from the image search results) and images that are identified as being relevant to another unique query. The trained model is applied to the content feature values of all known images to generate an image relevance score that can be used to order search results for the query.