So using a table-based approach to find the nearest color now, using only the upper 6 bits to stop the table from getting too large. It still takes noticeable time to build it. :/
But as a result, the video actually looks okay now...
svn-id: r35298
Since Indeo 3 uses full 24bit colors (in YUV-space), frame data is mapped to the current palette (using an octree, should be the fasted way, methinks) and dithered with the Sierra-2-4A ("Filter Light") error distribution algorithm.
svn-id: r35290