Convolutional dynamic auto-encoder: a clustering method for semantic images

Autoencoders have been employed in various deep embedded clustering methods, but they suffer from Feature Randomness and Feature Drift problems when it comes to high-semantic data. Dynamic autoencoder (DynAE) provides a better trade-off between Feature Randomness and Feature Drift, thanks to its dynamic objective function. The purpose of this study was to propose convolutional dynamic autoencoder (ConvDynAE) as a novel model for deep embedded clustering methods on highly semantic image datasets. ConvDynAE is an improved model of DynAE that is built with convolutional autoencoder architecture which replaces the traditional autoencoder. The proposed model consists of the pretraining phase and clustering phase. An experimental study was conducted on benchmark datasets, and results were compared with state-of-the-art methods, which showed the superiority of the proposed model .