Object Recognition in Images Using GAN-Based Appearance Synthesis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image recognition technologies struggle to accurately identify individuals due to changes in outward appearances such as pose, clothes, and hairstyle over time or in different surroundings.

Innovation Solution

An electronic device uses a learning model, such as a generative adversarial network (GAN), to generate a new image based on extracted object information, considering pose and style, and compares it with the original image to recognize the object accurately.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional image recognition methods are used to identify persons, then the system is simple and fast, but the identification accuracy decreases when outward appearance changes

Engineering Contradiction:
Improveidentification accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by generating multiple synthesized images showing the target person in various poses, hairstyles, and clothing styles before the actual identification process. This preprocessing step creates a comprehensive reference database that accounts for appearance variations, enabling accurate identification even when the captured image shows different outward features.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates synthetic copies of the target person's images through GAN-based generation, producing multiple variations that simulate different appearances. These copied images serve as reference patterns for comparison, allowing the system to identify the person regardless of changes in pose, hairstyle, or clothing in the actual captured image.

Inventive Principle:
Principle #26Copying

2Measurement precision

If multiple factors such as pose, clothes, and hairstyle are considered for accurate identification, then the identification accuracy improves, but the complexity of the recognition system increases

Engineering Contradiction:
Improveidentification accuracyVSAvoidrecognition system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the identification task into distinct components: pose detection, hairstyle recognition, and clothing analysis. Each component is processed separately through specialized neural network modules, allowing the system to handle multiple appearance factors independently and then integrate the results for comprehensive identification.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system introduces an intermediary GAN-based synthesis module that acts as a mediator between the captured image and the identification process. This intermediary generates synthetic reference images that bridge the gap between different appearances, enabling accurate comparison and identification without requiring the system to directly handle all appearance variations simultaneously.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Quantity of substance

If the system captures images at different times and surroundings, then more data is available for identification, but the variation in outward appearance makes recognition more difficult

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

Solution Approach 1:

The system deliberately changes key appearance parameters such as pose angles, hairstyle styles, and clothing types when generating synthetic reference images. By systematically varying these parameters to create diverse reference patterns, the system prepares for and accommodates the natural variations that occur when capturing images at different times and in different surroundings.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3879449B1Electronic device for recognizing object in image, and operation method thereof
Publication Date: 2025.10.29 SAMSUNG ELECTRONICS CO LTD
  • EP3879449B1 patent drawingFigure 1
  • EP3879449B1 patent drawingFigure 2
  • EP3879449B1 patent drawingFigure 3

AI summary

Provided is a method, performed by an electronic device, of recognizing an object included in an image, the method including: extracting first object information from a first object included in a first image, obtaining a learning model for generating an image including a second object from the first object information, generating a second image including the second object by inputting the first object information to the learning model, comparing the first image with the second image, and recognizing the first object as the second object in the first image, based on a result of the comparing.