Deep Learning based Image Classification using Auto-encoders

In this paper, a novel approach for image clustering has been proposed, that combines the power of Deep Learning Feature Extraction with traditional clustering techniques. The proposed method utilizes Auto-Encoders and MobileNetV2, a lightweight Convolutional Neural Network, as feature extractors to transform raw image data into compact and meaningful representations. These representations are subsequently clustered using the k-means algorithm. By leveraging the pre-trained MobileNetV2 and an autoencoder for dimensionality reduction, high-quality clustering can be achieved, while maintaining computational efficiency. The approach is validated on a large image dataset, demonstrating improved clustering accuracy and better intra-cluster cohesion as compared to traditional clustering methods applied directly to pixel data. This method opens new possibilities for scalable and efficient image clustering in large-scale visual datasets.