Abstract: A learning television system will transmit only true information data which bandwidth is not dependent on screen size or resolution but rather only on novelty and movement within the image. This invention will allow for broadcast transmission of High Definition Television (HDTV) using present TV channels. Before transmission the system is made to learn image pattern by simply pointing the television camera at familiar scenes. During transmission the input images are de-composed into a layered progression of partial images, called superpixel, by comparison with previously learned images. Only the moving portion of an image is transmitted as superpixel where each superpixel may represent any arbitrary large portion of the screen. Superpixel transmission provides great data compression in an all digital TV as well as a virtually unbreakable encryption code for secure communication. The system also provides for roboter vision in a self-learning computer system.

Abstract: A knowledge storage and retrieval method is performed to result in a single address number of a storage region which represents a large quantity of information. A first character of a data sequence and a starting number are directed into a two-position buffer to form an input matrix. The storage locations of the storage region are addressed to see if the matrix is found in the storage region. If the matrix is found, the address number of the storage location of the storage region at which it is found is then stored in the second position of the buffer region. Then the next data character is directed into the first position of the buffer region and the searching step is repeated. If the matrix in the buffer region is not found in the storage region, the matrix is stored in a free storage location of the storage region and the address number of the free storage location is directed into the second position of the buffer.