Image Training Set Analysis for Targeted Recognition Improvement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image classification systems face challenges in determining the most effective types of training data to improve their performance, as adding similar images may not significantly enhance accuracy, while diverse images with specific characteristics can provide greater gains.

Innovation Solution

An improvement model suggests the types of training images that will most enhance the image classification system's ability to recognize objects by analyzing feature characteristics such as facial expressions, accessories, and image quality metrics, using a random decision forest model to estimate performance gains.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If diverse training images with specific characteristics are added, then object recognition accuracy is improved, but training data selection complexity increases

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidtraining data selection complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system automatically analyzes the training set characteristics and identifies which types of images would improve performance without requiring manual expert analysis. The improvement model self-services by generating recommendations based on analyzing gaps between current training data characteristics and desired performance outcomes.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system uses performance metrics from validation images to provide feedback about current model capabilities, then uses this feedback to guide selection of additional training images that specifically address identified weaknesses in object recognition.

Inventive Principle:
Principle #23Feedback

2Quantity of substance

If multiple similar images are added to training set, then training data quantity increases, but performance gain is minimal

Engineering Contradiction:
Improvetraining data quantityVSAvoidrecognition accuracy
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

Instead of simply increasing quantity of training images, the system changes the parameters of the training data by selecting images with specific characteristics that differ from the current training set, such as different poses, lighting conditions, or object variations that address identified performance gaps.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The system identifies specific deficiencies in the training set (local quality issues) and adds images that specifically address those deficiencies rather than uniformly adding all types of images, thereby optimizing the quality composition of the training data.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3791318B1Training set sufficiency for image analysis
Publication Date: 2026.03.25 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3791318B1 patent drawingFigure 1
  • EP3791318B1 patent drawingFigure 2
  • EP3791318B1 patent drawingFigure 3

AI summary

Aspects of the technology described herein improve an object recognition system by specifying a type of picture that would improve the accuracy of the object recognition system if used to retrain the object recognition system. The technology described herein can take the form of an improvement model that improves an object recognition model by suggesting the types of training images that would improve the object recognition model's performance. For example, the improvement model could suggest that a picture of a person smiling be used to retrain the object recognition system. Once trained, the improvement model can be used to estimate a performance score for an image recognition model given the set characteristics of a set of training of images. The improvement model can then select a feature of an image, which if added to the training set, would cause a meaningful increase in the recognition system's performance.