| 摘要 |
A method for re-indexing a palette-indexed image is disclosed. The method uses an array of symbol cross-counts that indicate the degree of occurrence, within the image, of symbols in one or more predefined contextual relationships, such as symbol adjacency. One objective of the method is to manipulate the palette index such that adjacent symbols in the image are assigned indices that are as close as possible in symbol space, thus enhancing the subsequent compressability of the image with many lossless compressors. As global minimization is generally computationally impracticable, the disclosed embodiments present a greedy suboptimal solution to this problem.The basic method uses a one-dimensional reassignment pool and a seed symbol. A single symbol is selected for positioning either to the immediate right or left of the seed in the pool, according to a potential function that uses the cross-count array. This process is then iterated, considering the first and second pool symbols during the next selection, etc., placing symbols in the pool so as to minimize the average interpixel differences in the re-indexed image.
|
