A Comparative Study of the Application of Metamorphic Testing in Image Recognition and Processing

The rapid advancement of artificial intelligence (AI) and deep learning technologies has significantly improved image recognition systems, impacting critical domains like medical diagnostics, autonomous driving, and biometric security. However, validating these complex systems remains challenging du...

Descripción completa

Detalles Bibliográficos
Autor: Su, Bingyi
Tipo de recurso: tesis de maestría
Fecha de publicación:2025
País:España
Institución:Universidad Complutense de Madrid (UCM)
Repositorio:Docta Complutense
Idioma:inglés
OAI Identifier:oai:docta.ucm.es:20.500.14352/124879
Acceso en línea:https://hdl.handle.net/20.500.14352/124879
Access Level:acceso abierto
Palabra clave:004(043.3)
Software Testing
Oracle Problem
Metamorphic Testing
Test Automation
GeMTest
Image Classification
Pruebas de software
Problema del oráculo
Testing Metamórfico
Automatización de pruebas
Clasificación de imágenes
Informática (Informática)
33 Ciencias Tecnológicas
Descripción
Sumario:The rapid advancement of artificial intelligence (AI) and deep learning technologies has significantly improved image recognition systems, impacting critical domains like medical diagnostics, autonomous driving, and biometric security. However, validating these complex systems remains challenging due to the oracle problem, this is the difficulty in determining correct outputs when clear expected results are unavailable. This thesis explores metamorphic testing (MT), a testing method using relationships between inputs and outputs (metamorphic relations, MRs), to effectively address the oracle problem. Focusing on a practical evaluation, the study applies metamorphic testing to a ResNet- 50 classifier validated on the ImageNet-1K dataset. Experiments using various MRs were automated through the GeMTest framework. Results show metamorphic testing’s practical effectiveness in detecting classification errors while highlighting limitations such as high false positive rates. The thesis concludes by summarising practical insights and suggesting realistic future research directions.