Image Label Confidence Adjustment via Proximity Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image labeling methods are inefficient due to manual labeling being time-consuming and inconsistent, and computer vision techniques are limited in accuracy, leading to suboptimal image search results.
Innovation Solution
An image labeling system that accesses images with associated labels and confidence scores, calculates proximity scores based on metrics like time, location, and social connections, and adjusts label confidence scores to improve image relevance ranking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to associate labels with images, then label accuracy can be improved, but time consumption and inconsistency increase
Solution Approach 1:
The system enables images to automatically contribute to each other's labeling through the propagation mechanism. Each image's labels and confidence scores are used to adjust related images' labels, creating a self-service system where the collective image set improves individual image labeling without manual intervention for each image.
Solution Approach 2:
The system implements feedback by using the labels and confidence scores from one image to adjust the labels of related images. The propagation process creates a feedback loop where labeling information circulates through the image set, with higher confidence labels reinforcing related images' labels and improving overall consistency.
2Productivity
If computer vision techniques are used to automatically create labels, then time efficiency is improved, but label accuracy is limited
Solution Approach 1:
The system merges the outputs of multiple computer vision techniques by combining labels and confidence scores from multiple images. Instead of relying on a single computer vision system's limited accuracy, the propagation mechanism integrates information from multiple sources, allowing the collective result to exceed individual system limitations.
Solution Approach 2:
The system creates a universal labeling approach that works across different computer vision techniques and image types. The propagation mechanism is technique-agnostic, accepting labels from various computer vision systems and harmonizing them through the confidence score-based adjustment process, making the system universally applicable regardless of the specific computer vision technique used.
3Measurement precision
If labels are adjusted based on multiple image signals, then image search accuracy is improved, but system complexity increases
Solution Approach 1:
The system changes the parameter of confidence scores from static values to dynamic, adjustable parameters. By allowing confidence scores to be modified based on proximity scores and related images' labels, the system transforms fixed labeling data into flexible, adaptive parameters that improve search accuracy without requiring complex reprocessing of the entire image set.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for adjusting confidence scores of image labels for images. In one aspect, a method includes accessing images stored in an image data store, the images being associated with respective sets of labels, the labels describing content depicted in the image and having a respective confidence score that is a measure of confidence that the label accurately describes the content depicted in the image; selecting a first image from the images and determining for each of the other images and independent of the labels, a proximity score that is a measure of a relatedness of the other image to the first image; and adjusting the set of labels associated with the first image based on the respective proximity scores of the other images and the confidence scores of the labels of the other images.