Image Search Refinement Using Visual Facets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional electronic commerce systems require multiple searches and significant resource allocation for manual tagging to identify and rank items based on search criteria, leading to inefficiencies in presenting similar items to users.
Innovation Solution
An image search service generates facets such as color, texture, and shape from item images, creating descriptors that allow for automatic tagging and ranked similarity searches without manual intervention, enabling users to find similar items through a single search query.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual tagging process is used to create metadata database, then item attributes can be identified, but significant resources are consumed and the process is time-consuming
Solution Approach 1:
The patent replaces the manual mechanical tagging process with an automated image recognition system. The system uses computer vision algorithms to automatically analyze item images, extract visual features (color, texture, shape), and generate metadata descriptors without human intervention, thereby substituting manual labor with automated computational processes
Solution Approach 2:
The system enables self-service by allowing items to be automatically tagged through their own images. When items are uploaded to the electronic commerce system, the image recognition service automatically processes their images to extract attributes and create metadata, eliminating the need for separate manual tagging operations
2Ease of operation
If conventional search method is used to identify items, then items matching search criteria can be found, but multiple searches are required to rank items by similarity
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing visual feature descriptors for all items in the database during the automated tagging phase. This pre-processing creates a structured metadata database with extracted attributes (color histograms, texture features, shape descriptors) that enables rapid similarity comparison during search operations, eliminating the need for multiple sequential searches
Solution Approach 2:
The system merges the item identification and ranking functions into a single search operation. By using the pre-computed visual feature descriptors and similarity metrics, the search system can simultaneously identify matching items and rank them by similarity in one query, combining what were previously separate operations
3Loss of information
If metadata database is created and maintained manually, then item attributes are available for search, but significant resources are necessary for creation and maintenance
Solution Approach 1:
The patent replaces manual metadata creation and maintenance with automated image analysis. The system uses computer vision algorithms to continuously and automatically extract visual features from item images, generating and updating metadata descriptors without human intervention, thereby eliminating the labor-intensive manual processes while ensuring information availability
Data Source
AI summary
One exemplary embodiment involves generating a plurality of facets and descriptors of all items available via an electronic shop service based on the images depicting each of the items. The facets that may be generated include a color facet, a shape facet, a texture facet, and/or other facets. The embodiment further involves receiving search criteria from a user that includes a number of search criteria elements. Each element of the search criteria may be represented by an image. The embodiment involves generating a search criteria descriptor based on the images received in connection with the search criteria. Additionally, the embodiment involves identifying items associated with images that correspond with the search criteria descriptor and providing the images of the identified items in a user interface.