Image Recognition Candidate Clustering for Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition systems face inefficiencies in processing a large number of image match candidates, leading to increased computational expense and latency, as well as the potential for the ideal match to be missed due to geometric verification limitations.
Innovation Solution
The system clusters image match candidates into groups or equivalence classes at runtime, allowing for geometric processing of a representative image per group instead of each individual image, thereby reducing the number of images that need to be geometrically verified.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the maximum number of candidates is increased to ensure the ideal match is not missed, then the reliability of matching is improved, but the computational expense and latency increase significantly
Solution Approach 1:
The patent merges similar or near-duplicate images into equivalence classes or clusters, where each cluster represents a group of visually similar images. By processing one representative image per cluster instead of every individual image, the system maintains matching reliability while significantly reducing the number of geometric verifications required, thus decreasing query latency.
Solution Approach 2:
The patent segments the set of candidate images into multiple equivalence classes based on visual similarity. This segmentation allows the system to handle large numbers of candidates efficiently by dividing them into manageable groups, where only representative images from each group undergo expensive geometric processing.
2Measurement precision
If geometric processing is performed on every candidate image to prevent non-matching results, then the measurement precision is improved, but the computational complexity increases
Solution Approach 1:
The patent combines multiple similar candidate images into single equivalence classes, performing geometric processing on one representative image per class. This merging approach maintains matching precision by ensuring that if the representative image matches, all images in that equivalence class are considered matches, while dramatically reducing computational complexity by avoiding redundant processing of visually similar images.
3Reliability
If the number of candidates processed is increased to account for poor quality database images and distracting features, then the reliability is improved, but the processing time increases
Solution Approach 1:
The patent merges candidates into equivalence classes based on visual similarity metrics. This allows the system to efficiently handle increased numbers of candidates that arise from poor quality images and distracting features, maintaining reliability by processing representative images from each class while reducing overall processing duration through the reduced number of geometric verifications required.
Data Source
AI summary
Various embodiments enable an image recognition system reduce the number image match candidates before running a full-fledged pair-wise match on all image match candidates. In order to accomplish this, each inventory image can be assigned to a group. For example, a title for a book sold by an electronic marketplace could be available in multiple languages, in multiple bindings, and the book could be available in print, audio book, or electronic book. Each one of these variations could be associated with its own similarly looking inventory image, each of which could be returned as a valid match to a query image for the book. Accordingly, the inventory images for these variations could be assigned to a group for the book and, instead of geometrically processing an image for each variation, the image match system can process a single image representing all of the variations.