The work presented in this paper is in the general framework of classification using deep learning and, more precisely, that of convolutional Autoencoder. In particular, this last proposes an alternative for the processing of high-dimensional data, to facilitate their classification. In this paper, we propose the incorporation of convolutional autoencoders as a general unsupervised learning data dimension reduction method for creating robust and compressed feature representations for better storage and transmission to the classification process to improve K-means performance on image classification tasks. The experimental results on three image databases, MNIST, Fashion-MNIST, and CIFAR-10, show that the proposed method significantly outperforms deep clustering models in terms of clustering quality. Povzetek: Ta članek predlaga vključitev konvolucijskih samodejnih kodirnikov kot splošne nenadzorovane metode zmanjševanja razsežnosti podatkov za učenje izboljšanja zmogljivosti k-means algoritma pri klasifikaciji slik.