Huffman Coding with Non-Sorted Frequencies Shmuel Tomi Klein Dana Shapira Bar Ilan University, Ashkelon Academic College, ISRAEL
Outline Background and motivation Using non-sorted frequencies Relevance to other compression Conclusions Dynamic compression of data packets Background and motivation Using non-sorted frequencies Dynamic compression of data packets Relevance to other compression Conclusions
Sorted frequencies Sufficient but not necessary Huffman’s algorithm Construction in Code construction only
Applications Low text / code ratio Several codes Markov process encoding Dynamic coding schemes: Encoding based on
Optimal Trees Huffman Tree
Optimal Trees Non-Huffman Tree
Optimal Trees Non-Huffman Tree
No restriction bad encoding n2n n2n
Restriction: order operations
Start with any order Then use 2 queues Reversed order full tree
Partial Sort: parameter K W1W1 WKWK...W3W3 W2W2 Time: 4-grams3-grams2-grams1-gramsTests: English French
Average # bits/char vs # partition blocks English grams 2-grams 3-grams 4-grams
French Average # bits/char vs # partition blocks grams 2-grams 3-grams 4-grams
Dynamic compression of data packets Encoding based on
Bub- For-5 BubbleBlocked Block size Bigrams Comprs Time Comprs Time French English
Relevance to other compression schemes Arithmetic coding 256-ary Huffman (s,c)-dense codes, Fibonacci codes Burrows-Wheeler Transform
Conclusion Not fully sorting the weights Time savings for sort intensive methods Compresion / Time tradeoff