vp8 tree coding

VP8 tree coding is similar to Huffman coding algorithm which use pre-defined coding tree to decode (encode) syntax element value from (into) binary bit strings.

“A tree representing an encoding of an alphabet of n possible values always contains n-1 non-leaf nodes". In VP8, these coding trees are 2-forks trees stored in sequential arrays with positive items represent non-leaf nodes (sub-trees) while the non-positive items represent the leaf nodes.

Take ymode as example:

typedef enum
{
    DC_PRED,  
    V_PRED,   
    H_PRED,   
    TM_PRED,  
    B_PRED,   
    num_ymodes
}
intra_mbmode;

const tree_index ymode_tree [2 * (num_ymodes - 1)] =
{
  -DC_PRED, 2,         	/* root: DC_PRED = "0", "1" subtree */
  4, 6,               	/* "1" subtree has 2 descendant subtrees */
  -V_PRED, -H_PRED,    	/* "10" subtree:  V_PRED = "100", H_PRED = "101" */
  -TM_PRED, -B_PRED    	/* "11" subtree: TM_PRED = "110", B_PRED = "111" */
};

The ymode_tree can be represented as follow:

The tree decoding process is like this:

idx = 0;
while(){
    idx += read_one_bit();
    item = coding_tree[ idx ];
    if( item > 0 ) {
        idx = item;         // jump to subtree
    }else{
        return (-item);     // got a leaf
    }
}