| OctreeQuantizer |
The idea here is to build a tree structure containing always a maximum of K different colors. If a further color is to be added to the tree structure, its color value has to be merged with the most likely one that is already in the tree. The both values are substituted by their mean. The most important data structure are the nodes of the octree. Each inner node of the octree contain a maximum of eight successors, the leave nodes keep information for the color value (colorvalue), the color index (colorindex), and a counter (colorcount) for the pixel that are already mapped to a particular leave. Because each of the red, green and blue value is between 0 and 255 the maximum depth of the tree is eight. In level i Bit i of RGB is used as selector for the successors. The octree is constructed during reading the image that is to be quantized. Only that parts of the octree are created that are really needed. Initially the first K values are represented exactly (in level eight). When the number of leaves nodes (currentK) exceeds K, the tree has to reduced. That would mean that leaves at the largest depth are substituted by their predecessor. |