Image data compression using counterpropagation network

The counterpropagation network functions as a statistically optimal self-adapting look-up table. When using this network for image data compression the Kohonen network generates a series of vector class indices with the input of subimages that come from the orthogonally divided pictorial image. These indices along with the weight vectors of the outstar network which has learned the vectors associated with the classes can be stored for reconstruction of the original image. The learning of intermediate forms of vector classes, the compression process, and the results, such as the compression ratios and the distortion ratios with respect to the target data, the compression unit, and the restored image, are discussed.<>