Item Image Similarity Scoring for Reducing Duplicate Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online transaction systems like eBay face inefficiencies due to the large number of duplicate or similar item images, leading to increased memory usage and delayed search responses, as they maintain over a billion item images with many listings being identical or substantially similar.
Innovation Solution
A system and method that generate a similarity score for existing item images by comparing new item listings with existing ones, proposing the most similar images to the user, and allowing them to associate their listings with these images, thereby reducing redundant image storage and improving search efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the system maintains all uploaded item images, then complete image coverage for all listings is achieved, but memory usage increases significantly
Solution Approach 1:
The patent merges multiple identical or similar item images into a single representative image. When duplicate images are detected through similarity comparison, the system consolidates them by keeping one instance and eliminating redundant copies, thereby reducing memory usage while maintaining complete image coverage for all listings.
Solution Approach 2:
The system discards redundant duplicate images after identifying them through similarity scoring, while recovering or preserving the essential image information by maintaining a single representative copy. This allows the system to eliminate unnecessary data storage while retaining all unique visual content.
2Reliability
If the system stores all uploaded item images, then no image data is lost, but search response time is delayed
Solution Approach 1:
By merging duplicate images into single representative instances, the system reduces the total number of images that need to be processed during search operations. This consolidation maintains data completeness while significantly improving search response time by reducing the search space.
Solution Approach 2:
The system extracts and eliminates redundant duplicate images from the storage system, keeping only unique and representative images. This extraction process removes unnecessary data that would otherwise slow down search operations while preserving all essential image information.
3Measurement precision
If the system uploads unique images for each listing, then image accuracy is maximized, but the number of redundant images increases
Solution Approach 1:
The system merges images that are determined to be identical or substantially similar through similarity comparison, consolidating multiple instances into a single representative image. This merging process reduces the total number of images stored while maintaining image accuracy by preserving the original high-quality representative image.
4Reliability
If the system loads all images into memory, then complete image availability is ensured, but memory efficiency decreases
Solution Approach 1:
The system merges duplicate images into single representative instances, reducing the total memory footprint required to store and access image data. This consolidation ensures that all unique images remain available while significantly improving memory efficiency by eliminating redundant data storage.
Data Source
AI summary
Techniques for sharing item images based on a similarity score are described. For example, a machine receives a description of an item from a user device. The machine accesses an existing image. The existing image is associated with one or more existing descriptions of the item. The machine selects, based on a description selection rule, the one or more existing descriptions corresponding to the existing image for comparison with the description received from the user device. The machine generates a similarity score for the existing image based on the comparison between the description received from the user device and the one or more existing descriptions corresponding to the existing image. The machine causes a display, in a user interface of the user device, of a number of existing images having corresponding similarity scores that exceed a threshold value.