摘要 |
A new lossless compression technique is provided that is specifically designed for palettized synthetic images which typically consist of several sharply delineated uniform-color regions with overlaid text and symbols. The associated algorithm uses patterns of previously-coded neighboring pixels to predict and code each pixel. Preferably, the four immediate and previously-coded neighbors of the current pixel are used. That is, the pixel immediately to the left (or west), the pixel immediately above and diagonally to the left (or northwest), the pixel immediately above (or north), and the pixel immediately above and diagonally to the right (or northeast). The basic pattern determined by the neighboring pixels is augmented by incorporating the number of times the value of each of the neighboring pixels is repeated in the corresponding direction. The prediction rules for different patterns are learned and updated from the image itself during the coding and decoding processes.
|