The Impact of Robust Learning on Unsupervised Image Classification

Accurately classifying images is a crucial aspect of many image processing tasks. While supervised deep network structures are dominant in machine learning problems, they have limitations in certain cases. In consequence, unsupervised image classification approaches can be employed for specific tasks. Various unsupervised image classification methods exist with different approaches. However, concerns have been raised about the performance limitations of these approaches. Additionally, new performance-enhancing methods and their effectiveness require further investigation. In this study, robust learning techniques were applied to the outputs of a joint approach image classification model. Robust learning techniques improved the model’s calibration and increased its accuracy. Furthermore, different strategies for sample selection in the robust learning model were tested and compared.