Joan P Mitchell

US Patent:

46334904, Dec 30, 1986

Filed:

Mar 15, 1984

Appl. No.:

6/589981

Inventors:

Gerald Goertzel - White Plains NY
Joan L. Mitchell - Ossining NY

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

H04B 166

US Classification:

375122

Abstract:

Data compression for transfer (storage or communication) by a continuously adaptive probability decision model, closely approaches the compression entropy limit. Sender and receiver perform symmetrical compression/decompression of binary decision n according to probabilities calculated independently from the transfer sequence of 1. . . n-1 binary decisions. Sender and receiver dynamically adapt the model probabilities, as a cumulative function of previously presented decisions, for optimal compression/decompression. Adaptive models for sender and receiver are symmetrical, to preserve data identity; transfer optimization is the intent. The source model includes a state generator and an adaptive probability generator, which dynamically modify the coding of decisions according to state, probability and bit signals, and adapt for the next decision. The system calculates probability history for all decisions, including the current decision, but uses probability history for decision n-1 (the penultimately current decision) for encoding decision n (the dynamically current decision). The system, separately at source and destination, reconfigures the compression/expansion algorithm, up to decision n-1, codes each decision in the data stream optimally, according to its own character in relation to the calculated probability history, and dynamically optimizes the current decision according to the transfer optimum of the data stream previously transferred.

US Patent:

T9850058, Aug 7, 1979

Filed:

Oct 16, 1978

Appl. No.:

5/951532

Inventors:

Gerald Goertzel - White Plains NY
Joan L. Mitchell - Peekskill NY

International Classification:

G06F 1520

US Classification:

364515

Abstract:

a dual-mode encoding and decoding procedure enables image data to be compressed optionally in one-dimensional (1D) mode or two-dimensional (2D) mode. In 1D mode, color transitions in the image are encoded as run length features only. In 2D mode, the transitions are encoded as vertical correlation features wherever possible, and where this is not possible, the transitions are encoded as run length features. The compression achieved by run length encoding in 2D mode may be enhanced in those instances where the "history line" which precedes the current scan line contains a transition located between points that are vertically aligned with the beginning and end points of the run currently being encoded. Run length counting is suspended for those pels in the current run that could have been referenced to the history transition if the run had ended with any of these pels, thereby enabling the run to be encoded as though it contained fewer pels than its actual length. Compression may be enhanced still further by dynamically interchanging the variable-length bit patterns respectively representing certain vertical correlation and run length prefix codes depending upon whether the preceding transition was encoded as a vertical correlation feature or a run length feature.