Learning Data Generation Apparatus for Image Recognition

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Generating learning models for image recognition using deep learning requires a large amount of data, which is time-consuming and inefficient.

Innovation Solution

A learning data generation method that combines image data from two sources based on specific positional relationships between regions of interest, using a processor to generate new image and ground truth data by aligning regions separated by a threshold distance, thereby reducing the amount of data needed for training.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a large amount of learning data is used for deep learning, then recognition accuracy is improved, but training time increases significantly

Engineering Contradiction:

Improverecognition accuracyVSAvoidtraining time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent combines multiple image data sets with different regions of interest to create a new learning data set. By merging images from different sources and applying spatial transformations (translation, rotation, flipping) to the second image data, the system generates diverse training samples that improve recognition accuracy while reducing the total amount of data needed, thereby shortening training time.

Inventive Principle:

Principle #5Merging (Combining)

2Adaptability or versatility

If more learning data is generated by combining images, then variations of learning data are increased, but the complexity of data processing increases

Engineering Contradiction:

Improvevariations of learning dataVSAvoiddata processing complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent divides the image processing into distinct segments: first image data is processed to extract regions of interest, second image data undergoes spatial transformations, and then the processed segments are combined. This segmentation of the data processing pipeline makes the complex operation more manageable and systematic, reducing processing complexity while maintaining data variation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies dynamic spatial transformations to the second image data, including translation by predetermined distances, rotation by predetermined angles, and flipping. These dynamic transformations generate diverse variations of learning data without requiring complex manual data processing, thereby increasing adaptability while keeping processing complexity manageable.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS20240303974A1Learning data generation apparatus and method, and learning model generation apparatus and method

Publication Date: 2024.09.12 FUJIFILM CORP

AI summary

There are provided a learning data generation apparatus and method and a learning model generation apparatus and method that can attain efficient learning. The learning data generation apparatus acquires first image data and second image data each having a region of interest, and when a positional relationship between the region of interest of the first image data and the region of interest of the second image data satisfies a predetermined condition, combines an image of a region, of the first image data, that includes the region of interest and an image of a region, of the second image data, that includes the region of interest to generate third image data. The learning model generation apparatus acquires the third image data generated by the learning data generation apparatus and trains a learning model by using the third image data to generate the learning model.