lucas roguski

projects repository

September 27th, 2011

This project focused on neural networks, exactly on using Kohonen’s self–organising maps for image compression.

The image compression is generally divided into two steps. Firstly, a neural network is being created and trained according to provided input images — the neurons are “trained” to recognize or react on specified block patterns. Secondly, image is being compressed — chunks of image data are being mapped to neurons, so that blocks of image data are described by neurons.

Application was written in Python (to try something new). The main algorithm was designed by me, where Tomasz ‘Tobek’ Chaja created GUI. Below are some sample images before and after compression, using different parameters when creating a neural map.

Result #1

Result #2

Comments are closed.