next up previous contents
Next: 4.4.1 Windows from natural Up: 4. Abstract recurrent neural Previous: 4.3 Function approximation on

4.4 Image Completion

The abstract RNN can be used to complete partially occluded images. The completion was first tested on small windows cut out from natural backgrounds, and second tested on human faces. To use the recall mechanism, the gray-scaled images need to be converted into data points (vectors). This conversion was achieved by writing an image row by row into a vector (pixel values scaled between 0 and 1).


Heiko Hoffmann