A Little Huffman Coding with Java Tricks

Monday, May 16, 2005

A Huffman code is a way to utilize a binary tree to construct a minimal-length encoding for messages where certain characters or groups of characters have known frequencies. The tree used for such an operation called a Huffman tree. Huffman codes are the most efficient compression method for random data and are often found as steps in other compression algorithms such as JPEG and Deflate (ZIP). Building such trees is a very common exercise for Computer Science and Math classes so I can skip the details.

To construct a Huffman tree, first initialize a series of nodes, which are simply data structures holding the represented character and its frequency within the message. The series is sorted by frequency and the two lowest frequency nodes are picked from the list and combined into a composite node, which is given the sum of its child node frequencies as its frequency. The composite node is added back to the sorted list and the process is repeated until there is only one node in the list, which is the head of the Huffman tree.

In Java, the treeset data structure maintains a sorted list of comparable objects and can be used for an implementation of the Huffman code algorithm in the last paragraph. One caveat is that a treeset is a set, which means that duplicate nodes will simply disappear. The work around is to rig the compareTo method, which is implemented by all comparable objects, so that equality is never returned. In this implementation getCount returns the frequency of the node.

public int compareTo(Object a) {
HuffmanNode ha = (HuffmanNode) a;
  if (this.getCount() == ha.getCount()) {
     return -1;
  } else {
     return this.getCount() - ha.getCount();
  }
} // end compareTo

public int buildTree() {
tsNodes.clear();
tsNodes.addAll(htNodes.values());

while (tsNodes.size() > 1) {
 Iterator it = tsNodes.iterator();
 HuffmanNode a = (HuffmanNode)it.next();
 it.remove();
 HuffmanNode b = (HuffmanNode)it.next();
 it.remove();

 // Combine node a and node b into a composite node
 HuffmanNode combinedNode = new HuffmanNode(a, b);
 tsNodes.add(combinedNode);
} // end while tree > 1

huffmanTreeRoot = (HuffmanNode) tsNodes.first();
...
} // end buildTree

The compression rates achieved by this simple algorithm are pretty good even though they are not nearly as good as more specific ones in common use, such as JPEG, gzip, b2zip, etc.