A new high speed and flexible interactive color image quantization method is proposed in this paper. Methods have been developed for color image quantization into over 256 levels. These methods, however, can not provide effective color quantization into the much smaller numbers (i. e., 8 or 16) of levels, nor flexible color selection reflecting a designer's feelings. The proposed method is intended to overcome there two problems using one selection parameter (D). This parameter D represents a maximum distance between representative and candidate color points in the R G B space. With smaller D-value, impressive color even in a smaller area can be preserved. With larger D-value, subtle color change can be expressed. An efficient multi-dimensional data structure, BD-tree, is introduced to realize high-speed and flexible quantization. Experiments with many standard color images reveal the flexibibility and superiority of the proposed method.
View full abstract