BLZW Compression Java

Uses Sais.java with dictionary persist, and initialisation corrections. Also with alignment fix and unused function removed. A 32 bucket context provides an effective 17 bit dictionary key, using just 12 bits, along with the BWT redundancy model. This should provide superior compression of text. Now includes the faster skip decode. Feel free to donate to grow some open source based on data compression and related codecs.





/* BWT/LZW fast wide dictionary. (C)2016-2017 K Ring Technologies Ltd.
The context is used to make 32 dictionary spaces for 128k symbols max.
This then givez 12 bit tokens for an almost effective 16 bit dictionary.
For an approximate 20% data saving above regular LZW.

The process is optimized for L2 cache sizea.

A mod 16 gives DT and EU collisions on hash.
A mod 32 is ASCII proof, and hence good for text.

The count compaction includes a skip code for efficient storage.
The dictionary persists over the stream for good running compression.
64k blocks are used for fast BWT. Larger blocks would give better
compression, but be slower. The main loss is the count compactio storage.

An arithmetic coder post may be effective but would be slow. Dictonary
acceleration would not necessarily be useful, and problematic after the
stream start. A 12 bit code is easy to pack, keeps the dictionary small
and has the sweet spot of redundancy in while not making large rare or
single use symbols.
*/

package uk.co.kring.net;

import java.io.EOFException;
import java.io.Externalizable;
import java.io.FilterInputStream;
import java.io.FilterOutputStream;
import java.io.IOException;
import java.io.ObjectInput;
import java.io.ObjectOutput;
import java.util.HashMap;

/**
 * Created by user on 06/06/2016.
 */
public class Packer {

    public static class OutputStream extends FilterOutputStream implements Externalizable {

        byte[] buf = new byte[4096 * 16];//64K block max
        int cnt = 0;//pointer to end
        int[] dmax = new int[32];
        HashMap<String, Integer> dict;

        public OutputStream(java.io.OutputStream out) {
            super(out);
        }

        @Override
        public void readExternal(ObjectInput input) throws IOException, ClassNotFoundException {
            out = (java.io.OutputStream)input.readObject();
            input.read(buf);
            cnt = input.readChar();
            dict = (HashMap<String, Integer>)input.readObject();
        }

        @Override
        public void writeExternal(ObjectOutput output) throws IOException {
            output.writeObject(out);
            output.write(buf);
            output.writeChar(cnt);
            output.writeObject(dict);
        }

        @Override
        public void close() throws IOException {
            flush();
            out.close();
        }

        private byte pair = 0;
        private boolean two = false;

        private void outputCount(int num, boolean small, boolean tiny) throws IOException {
            if(tiny) {
                out.write((byte)num);
                return;
            }
            if(small) {
                out.write((byte)num);
                pair = (byte)((pair << 4) + (num >> 8));
                if(two) {
                    two = false;
                    out.write(pair);
                } else {
                    two = true;
                }
                return;
            }
            out.write((byte)(num >> 8));
            out.write((byte)num);
        }

        @Override
        public void flush() throws IOException {
            outputCount(cnt, false, false);//just in case length
            char[] count = new char[256];
            if(dict == null) {
                dict = new HashMap<>();
                for(int i = 0; i < 32; i++) {
                    dmax[i] = 256;//dictionary max
                }
            }
            for(int i = 0; i < cnt; i++) {
                count[buf[i]]++;
            }
            char skip = 0;
            boolean first = true;
            char acc = 0;
            char[] start = new char[256];
            for(int j = 0; j < 2; j++) {
                for (int i = 0; i < 256; i++) {
                    if(j == 0) {
                        acc += count[i];
                        start[i] = acc;
                    }
                    if (count[i] == 0) {
                        skip++;
                        if (first) {
                            outputCount(0, false, true);
                            first = false;
                        }
                    } else {
                        if (skip != 0) {
                            outputCount(skip, false, true);
                            skip = 0;
                            first = true;
                        }
                        outputCount(count[i], false, true);
                        count[i] >>= 8;
                    }
                }
                if(skip != 0) outputCount(skip, false, true);//final skip
            }
            int[] ptr = new int[buf.length];
            byte[] bwt = new byte[buf.length];

            outputCount(Sais.bwtransform(buf, bwt, ptr, cnt), false, false);

            //now an lzw
            String sym = "";
            char context = 0;
            char lastContext = 0;
            int test = 0;
            for(int j = 0; j < cnt; j++) {
                while(j >= start[context]) context++;
                if(lastContext == context) {
                    sym += bwt[j];//add a char
                } else {
                    lastContext = context;
                    outputCount(test, true, false);
                    sym = "" + bwt[j];//new char
                }
                if(sym.length() == 1) {
                    test = (int)sym.charAt(0);
                } else {
                    if(dict.containsKey(context + sym)) {
                        test = dict.get(context + sym);
                    } else {
                        outputCount(test, true, false);
                        if (dmax[context & 0x1f] < 0x1000) {//context limit
                            dict.put(context + sym, dmax[context & 0x1f]);
                            dmax[context & 0x1f]++;
                        }
                        sym = "" + bwt[j];//new symbol
                    }
                }
            }
            outputCount(test, true, false);//last match
            if(!two) outputCount(0, true, false);//aligned data
            out.flush();
            cnt = 0;//fill next buffer
        }

        @Override
        public void write(int oneByte) throws IOException {
            if(cnt == buf.length) flush();
            buf[cnt++] = (byte)oneByte;
        }
    }

    public static class InputStream extends FilterInputStream implements Externalizable {

        @Override
        public void readExternal(ObjectInput input) throws IOException, ClassNotFoundException {
            in = (java.io.InputStream)input.readObject();
            input.read(buf);
            idx = input.readChar();
            cnt = input.readChar();
            dict = (HashMap<Integer, String>)input.readObject();
        }

        @Override
        public void writeExternal(ObjectOutput output) throws IOException {
            output.writeObject(in);
            output.write(buf);
            output.writeChar(idx);
            output.writeChar(cnt);
            output.writeObject(dict);
        }

        //SEE MIT LICENCE OF Sais.java

        private static void unbwt(byte[] T, byte[] U, int[] LF, int n, int pidx) {
            int[] C = new int[256];
            int i, t;
            //for(i = 0; i < 256; ++i) { C[i] = 0; }//Java
            for(i = 0; i < n; ++i) { LF[i] = C[(int)(T[i] & 0xff)]++; }
            for(i = 0, t = 0; i < 256; ++i) { t += C[i]; C[i] = t - C[i]; }
            for(i = n - 1, t = 0; 0 <= i; --i) {
                t = LF[t] + C[(int)((U[i] = T[t]) & 0xff)];
                t += (t < pidx) ? 1 : 0;
            }
        }

        byte[] buf = new byte[4096 * 16];//64K block max
        int cnt = 0;//pointer to end
        int idx = 0;
        int[] dmax = new int[32];
        HashMap<Integer, String> dict;

        private boolean two = false;
        private int vala = 0;
        private int valb = 0;

        private int reader() throws IOException {
            int i = in.read();
            if(i == -1) throw new EOFException("End Of Stream");
            return i;
        }

        private char inCount(boolean small, boolean tiny) throws IOException {
            if(tiny) return (char)reader();
            if(small) {
                if(!two) {
                    vala = reader();
                    valb = reader();
                    int valc = reader();
                    vala += (valc << 4) & 0xf00;
                    valb += (valc << 8) & 0xf00;
                    two = true;
                } else {
                    vala = valb;
                    two = false;
                }
                return (char)vala;
            }
            int val = reader() << 8;
            val += reader();
            return (char)val;
        }

        public InputStream(java.io.InputStream in) {
            super(in);
        }

        @Override
        public int available() throws IOException {
            return cnt - idx;
        }

        @Override
        public void close() throws IOException {
            in.close();
        }

        private void doReads() throws IOException {
            if(available() == 0) {
                two = false;//align
                if(dict == null) {
                    dict = new HashMap<>();
                    for(int i = 0; i < 32; i++) {
                        dmax[i] = 256;
                    }
                }
                cnt = inCount(false, false);
                char[] count = new char[256];
                char tmp;
                for(int j = 0; j < 2; j++) {
                    for (int i = 0; i < 256; i++) {
                        count[i] += tmp = (char)(inCount(false, true) << (j == 1?8:0));
                        if (tmp == 0) {
                            i += inCount(false, true) - 1;
                        }
                    }
                }
                for(int i = 1; i < 256; i++) {
                    count[i] += count[i - 1];//accumulate
                }
                if(cnt != count[255]) throw new IOException("Bad Input Check (character count)");
                int choose = inCount(false, false);//read index
                if(cnt < choose) throw new IOException("Bad Input Check (selected row)");
                byte[] build;//make this
                //then lzw
                //rosetta code
                int context = 0;
                int lastContext = 0;
                String w = "" + inCount(true, false);
                StringBuilder result = new StringBuilder(w);
                while (result.length() < cnt) {//not yet complete
                    char k = inCount(true, false);
                    String entry;
                    while(result.length() > count[context]) {
                        context++;//do first
                        if (context > 255)
                            throw new IOException("Bad Input Check (character count)");
                    }
                    if(k < 256)
                        entry = "" + k;
                    else if (dict.containsKey(((context & 0x1f) << 16) + k))
                        entry = dict.get(((context & 0x1f) << 16) + k);
                    else if (k == dmax[context & 0x1f])
                        entry = w + w.charAt(0);
                    else
                        throw new IOException("Bad Input Check (token: " + k + ")");
                    result.append(entry);
                    // Add w+entry[0] to the dictionary.
                    if(lastContext == context) {
                        if (dmax[context & 0x1f] < 0x1000) {
                            dict.put(((context & 0x1f) << 16) +
                                    (dmax[context & 0x1f]++),
                                    w + entry.charAt(0));
                        }
                        w = entry;
                    } else {
                        //context change
                        context = lastContext;
                        //and following context should be a <256 ...
                        if(result.length() < cnt) {
                            w = "" + inCount(true, false);
                            result.append(w);
                        }
                    }
                }
                build = result.toString().getBytes();
                //working buffers
                int[] wrk = new int[buf.length];
                unbwt(build, buf, wrk, buf.length, choose);
                idx = 0;//ready for reads
                if(!two) inCount(true, false);//aligned data
            }
        }

        @Override
        public int read() throws IOException {
            try {
                doReads();
                int x = buf[idx++];
                doReads();//to prevent avail = 0 never access
                return x;
            } catch(EOFException e) {
                return -1;
            }
        }

        @Override
        public long skip(long byteCount) throws IOException {
            long i;
            for(i = 0; i < byteCount; i++)
                if(read() == -1) break;
            return i;
        }

        @Override
        public boolean markSupported() {
            return false;
        }

        @Override
        public synchronized void reset() throws IOException {
            throw new IOException("Mark Not Supported");
        }
    }
}

Author: Jacko

Technical. Well is mass information conservation the reason for dark energy via uncertain geometry and photon exchange? Is dark matter conservation of acceleration with a gradient field heavy graviton? Does the KODEK work yet?

7 thoughts on “BLZW Compression Java”

  1. It’s close to working. A number of bugs have been fixed, and it’s down to perhaps a few sillies. There is an Externalizable interface, and some possible synchronized words need adding.

  2. You’ll likely want to use algorithmic memory and a dynamic dictionary entry construction system for optimal big O space requirements.

  3. The decompression context would allow the context to be extended well beyond 8-bits, the block size would have to be ramped up, and so the complexity might not be worth it unless going for world records. The context at some point will reintroduce a requirement to store more context segment length pointers offsetting the self partition mutual information gain.

Comments are closed.