Supervised learning as a rule demands to mark off the objects 
in the images and to give them the names before beginning the training process. Semisupervised learning process starts from the image clustering and only after that the names are assigned to the clusters that correspond to different objects. Semisupervised learning permits minimization of the time needed to prepare image databases for training process. In this paper we propose to use Hebbian ensemble neural networks for image clustering and further information coding. 
We suppose that this method will be useful for face recognition, 
texture recognition, handwritten text recognition 
and some other pattern recognition tasks.