unipdf/pdf/core/encoding.go

/*
 * This file is subject to the terms and conditions defined in
 * file 'LICENSE.md', which is part of this source code package.
 */

package core

// Implement encoders for PDF. Currently supported:
// - Raw (Identity)
// - FlateDecode
// - LZW
// - ASCII Hex

import (
	"bytes"
	"compress/zlib"
	"encoding/hex"
	"errors"
	"fmt"
	goimage "image"
	gocolor "image/color"
	"image/jpeg"
	"io"

	// Need two slightly different implementations of LZW (EarlyChange parameter).
	lzw0 "compress/lzw"

	lzw1 "golang.org/x/image/tiff/lzw"

	"github.com/unidoc/unidoc/common"
)

const (
	StreamEncodingFilterNameFlate    = "FlateDecode"
	StreamEncodingFilterNameLZW      = "LZWDecode"
	StreamEncodingFilterNameDCT      = "DCTDecode"
	StreamEncodingFilterNameASCIIHex = "ASCIIHexDecode"
	StreamEncodingFilterNameASCII85  = "ASCII85Decode"
)

type StreamEncoder interface {
	GetFilterName() string
	MakeDecodeParams() PdfObject
	MakeStreamDict() *PdfObjectDictionary

	EncodeBytes(data []byte) ([]byte, error)
	DecodeBytes(encoded []byte) ([]byte, error)
	DecodeStream(streamObj *PdfObjectStream) ([]byte, error)
}

// Flate encoding.
type FlateEncoder struct {
	Predictor        int
	BitsPerComponent int
	// For predictors
	Columns int
	Colors  int
}

// Make a new flate encoder with default parameters, predictor 1 and bits per component 8.
func NewFlateEncoder() *FlateEncoder {
	encoder := &FlateEncoder{}

	// Default (No prediction)
	encoder.Predictor = 1

	// Currently only supporting 8.
	encoder.BitsPerComponent = 8

	encoder.Colors = 1
	encoder.Columns = 1

	return encoder
}

// Set the predictor function.  Specify the number of columns per row.
// The columns indicates the number of samples per row.
// Used for grouping data together for compression.
func (this *FlateEncoder) SetPredictor(columns int) {
	// Only supporting PNG sub predictor for encoding.
	this.Predictor = 11
	this.Columns = columns
}

func (this *FlateEncoder) GetFilterName() string {
	return StreamEncodingFilterNameFlate
}

func (this *FlateEncoder) MakeDecodeParams() PdfObject {
	if this.Predictor > 1 {
		decodeParams := PdfObjectDictionary{}
		decodeParams["Predictor"] = MakeInteger(int64(this.Predictor))

		// Only add if not default option.
		if this.BitsPerComponent != 8 {
			decodeParams["BitsPerComponent"] = MakeInteger(int64(this.BitsPerComponent))
		}
		if this.Columns != 1 {
			decodeParams["Columns"] = MakeInteger(int64(this.Columns))
		}
		if this.Colors != 1 {
			decodeParams["Colors"] = MakeInteger(int64(this.Colors))
		}
		return &decodeParams
	}
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
// Has the Filter set and the DecodeParms.
func (this *FlateEncoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())

	decodeParams := this.MakeDecodeParams()
	if decodeParams != nil {
		dict["DecodeParms"] = decodeParams
	}

	return &dict
}

// Create a new flate decoder from a stream object, getting all the encoding parameters
// from the DecodeParms stream object dictionary entry.
func newFlateEncoderFromStream(streamObj *PdfObjectStream, decodeParams *PdfObjectDictionary) (*FlateEncoder, error) {
	encoder := NewFlateEncoder()

	encDict := streamObj.PdfObjectDictionary
	if encDict == nil {
		// No encoding dictionary.
		return encoder, nil
	}

	// If decodeParams not provided, see if we can get from the stream.
	if decodeParams == nil {
		obj := (*encDict)["DecodeParms"]
		if obj != nil {
			dp, isDict := obj.(*PdfObjectDictionary)
			if !isDict {
				common.Log.Debug("Error: DecodeParms not a dictionary (%T)", obj)
				return nil, fmt.Errorf("Invalid DecodeParms")
			}
			decodeParams = dp
		}
	}
	if decodeParams == nil {
		// Can safely return here if no decode params, as the following depend on the decode params.
		return encoder, nil
	}

	common.Log.Trace("decode params: %s", decodeParams.String())
	obj, has := (*decodeParams)["Predictor"]
	if !has {
		common.Log.Debug("Error: Predictor missing from DecodeParms - Continue with default (1)")
	} else {
		predictor, ok := obj.(*PdfObjectInteger)
		if !ok {
			common.Log.Debug("Error: Predictor specified but not numeric (%T)", obj)
			return nil, fmt.Errorf("Invalid Predictor")
		}
		encoder.Predictor = int(*predictor)
	}

	// Bits per component.  Use default if not specified (8).
	obj, has = (*decodeParams)["BitsPerComponent"]
	if has {
		bpc, ok := obj.(*PdfObjectInteger)
		if !ok {
			common.Log.Debug("ERROR: Invalid BitsPerComponent")
			return nil, fmt.Errorf("Invalid BitsPerComponent")
		}
		encoder.BitsPerComponent = int(*bpc)
	}

	if encoder.Predictor > 1 {
		// Columns.
		encoder.Columns = 1
		obj, has = (*decodeParams)["Columns"]
		if has {
			columns, ok := obj.(*PdfObjectInteger)
			if !ok {
				return nil, fmt.Errorf("Predictor column invalid")
			}

			encoder.Columns = int(*columns)
		}

		// Colors.
		// Number of interleaved color components per sample (Default 1 if not specified)
		encoder.Colors = 1
		obj, has = (*decodeParams)["Colors"]
		if has {
			colors, ok := obj.(*PdfObjectInteger)
			if !ok {
				return nil, fmt.Errorf("Predictor colors not an integer")
			}
			encoder.Colors = int(*colors)
		}
	}

	return encoder, nil
}

func (this *FlateEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	common.Log.Trace("FlateDecode bytes")

	bufReader := bytes.NewReader(encoded)
	r, err := zlib.NewReader(bufReader)
	if err != nil {
		common.Log.Debug("Decoding error %v\n", err)
		common.Log.Debug("Stream (%d) % x", len(encoded), encoded)
		return nil, err
	}
	defer r.Close()

	var outBuf bytes.Buffer
	outBuf.ReadFrom(r)

	common.Log.Trace("En: % x\n", encoded)
	common.Log.Trace("De: % x\n", outBuf.Bytes())

	return outBuf.Bytes(), nil
}

// Decode a FlateEncoded stream object and give back decoded bytes.
func (this *FlateEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	// TODO: Revamp this support to handle TIFF predictor (2).
	// Also handle more filter bytes and support more values of BitsPerComponent.

	common.Log.Trace("FlateDecode stream")
	common.Log.Trace("Predictor: %d", this.Predictor)
	if this.BitsPerComponent != 8 {
		return nil, fmt.Errorf("Invalid BitsPerComponent (only 8 supported)")
	}

	outData, err := this.DecodeBytes(streamObj.Stream)
	if err != nil {
		return nil, err
	}
	common.Log.Trace("En: % x\n", streamObj.Stream)
	common.Log.Trace("De: % x\n", outData)

	if this.Predictor > 1 {
		if this.Predictor == 2 { // TIFF encoding: Needs some tests.
			common.Log.Trace("Tiff encoding")
			common.Log.Trace("Colors: %d", this.Colors)

			rowLength := int(this.Columns) * this.Colors
			rows := len(outData) / rowLength
			if len(outData)%rowLength != 0 {
				common.Log.Debug("ERROR: TIFF encoding: Invalid row length...")
				return nil, fmt.Errorf("Invalid row length (%d/%d)", len(outData), rowLength)
			}

			if rowLength%this.Colors != 0 {
				return nil, fmt.Errorf("Invalid row length (%d) for colors %d", rowLength, this.Colors)
			}
			common.Log.Trace("inp outData (%d): % x", len(outData), outData)

			pOutBuffer := bytes.NewBuffer(nil)

			// 0-255  -255 255 ; 0-255=-255;
			for i := 0; i < rows; i++ {
				rowData := outData[rowLength*i : rowLength*(i+1)]
				// Predicts the same as the sample to the left.
				// Interleaved by colors.
				for j := this.Colors; j < rowLength; j++ {
					rowData[j] = byte(int(rowData[j]+rowData[j-this.Colors]) % 256)
				}
				pOutBuffer.Write(rowData)
			}
			pOutData := pOutBuffer.Bytes()
			common.Log.Trace("POutData (%d): % x", len(pOutData), pOutData)
			return pOutData, nil
		} else if this.Predictor >= 10 && this.Predictor <= 15 {
			common.Log.Trace("PNG Encoding")
			// Columns represents the number of samples per row; Each sample can contain multiple color
			// components.
			rowLength := int(this.Columns*this.Colors + 1) // 1 byte to specify predictor algorithms per row.
			rows := len(outData) / rowLength
			if len(outData)%rowLength != 0 {
				return nil, fmt.Errorf("Invalid row length (%d/%d)", len(outData), rowLength)
			}

			pOutBuffer := bytes.NewBuffer(nil)

			common.Log.Trace("Predictor columns: %d", this.Columns)
			common.Log.Trace("Length: %d / %d = %d rows", len(outData), rowLength, rows)
			prevRowData := make([]byte, rowLength)
			for i := 0; i < rowLength; i++ {
				prevRowData[i] = 0
			}

			for i := 0; i < rows; i++ {
				rowData := outData[rowLength*i : rowLength*(i+1)]

				fb := rowData[0]
				switch fb {
				case 0:
					// No prediction. (No operation).
				case 1:
					// Sub: Predicts the same as the sample to the left.
					for j := 2; j < rowLength; j++ {
						rowData[j] = byte(int(rowData[j]+rowData[j-1]) % 256)
					}
				case 2:
					// Up: Predicts the same as the sample above
					for j := 1; j < rowLength; j++ {
						rowData[j] = byte(int(rowData[j]+prevRowData[j]) % 256)
					}
				default:
					common.Log.Debug("ERROR: Invalid filter byte (%d) @row %d", fb, i)
					return nil, fmt.Errorf("Invalid filter byte (%d)", fb)
				}

				for i := 0; i < rowLength; i++ {
					prevRowData[i] = rowData[i]
				}
				pOutBuffer.Write(rowData[1:])
			}
			pOutData := pOutBuffer.Bytes()
			return pOutData, nil
		} else {
			common.Log.Debug("ERROR: Unsupported predictor (%d)", this.Predictor)
			return nil, fmt.Errorf("Unsupported predictor (%d)", this.Predictor)
		}
	}

	return outData, nil
}

// Encode a bytes array and return the encoded value based on the encoder parameters.
func (this *FlateEncoder) EncodeBytes(data []byte) ([]byte, error) {
	if this.Predictor != 1 && this.Predictor != 11 {
		return nil, fmt.Errorf("FlateEncoder Predictor = 1, 11 only supported")
	}

	if this.Predictor == 11 {
		// The length of each output row in number of samples.
		// N.B. Each output row has one extra sample as compared to the input to indicate the
		// predictor type.
		rowLength := int(this.Columns)
		rows := len(data) / rowLength
		if len(data)%rowLength != 0 {
			common.Log.Error("Invalid column length")
			return nil, errors.New("Invalid row length")
		}

		pOutBuffer := bytes.NewBuffer(nil)

		tmpData := make([]byte, rowLength)

		for i := 0; i < rows; i++ {
			rowData := data[rowLength*i : rowLength*(i+1)]

			// PNG SUB method.
			// Sub: Predicts the same as the sample to the left.
			tmpData[0] = rowData[0]
			for j := 1; j < rowLength; j++ {
				tmpData[j] = byte(int(rowData[j]-rowData[j-1]) % 256)
			}

			pOutBuffer.WriteByte(1) // sub method
			pOutBuffer.Write(tmpData)
		}

		data = pOutBuffer.Bytes()
	}

	var b bytes.Buffer
	w := zlib.NewWriter(&b)
	w.Write(data)
	w.Close()

	return b.Bytes(), nil
}

// LZW encoding/decoding functionality.
type LZWEncoder struct {
	Predictor        int
	BitsPerComponent int
	// For predictors
	Columns int
	Colors  int
	// LZW algorithm setting.
	EarlyChange int
}

// Make a new LZW encoder with default parameters.
func NewLZWEncoder() *LZWEncoder {
	encoder := &LZWEncoder{}

	// Default (No prediction)
	encoder.Predictor = 1

	// Currently only supporting 8.
	encoder.BitsPerComponent = 8

	encoder.Colors = 1
	encoder.Columns = 1
	encoder.EarlyChange = 1

	return encoder
}

func (this *LZWEncoder) GetFilterName() string {
	return StreamEncodingFilterNameLZW
}

func (this *LZWEncoder) MakeDecodeParams() PdfObject {
	if this.Predictor > 1 {
		decodeParams := PdfObjectDictionary{}
		decodeParams["Predictor"] = MakeInteger(int64(this.Predictor))

		// Only add if not default option.
		if this.BitsPerComponent != 8 {
			decodeParams["BitsPerComponent"] = MakeInteger(int64(this.BitsPerComponent))
		}
		if this.Columns != 1 {
			decodeParams["Columns"] = MakeInteger(int64(this.Columns))
		}
		if this.Colors != 1 {
			decodeParams["Colors"] = MakeInteger(int64(this.Colors))
		}
		return &decodeParams
	}
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
// Has the Filter set and the DecodeParms.
func (this *LZWEncoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())

	decodeParams := this.MakeDecodeParams()
	if decodeParams != nil {
		dict["DecodeParms"] = decodeParams
	}

	dict["EarlyChange"] = MakeInteger(int64(this.EarlyChange))

	return &dict
}

// Create a new LZW encoder/decoder from a stream object, getting all the encoding parameters
// from the DecodeParms stream object dictionary entry.
func newLZWEncoderFromStream(streamObj *PdfObjectStream, decodeParams *PdfObjectDictionary) (*LZWEncoder, error) {
	// Start with default settings.
	encoder := NewLZWEncoder()

	encDict := streamObj.PdfObjectDictionary
	if encDict == nil {
		// No encoding dictionary.
		return encoder, nil
	}

	// If decodeParams not provided, see if we can get from the stream.
	if decodeParams == nil {
		obj := (*encDict)["DecodeParms"]
		if obj != nil {
			dp, isDict := obj.(*PdfObjectDictionary)
			if !isDict {
				common.Log.Debug("Error: DecodeParms not a dictionary (%T)", obj)
				return nil, fmt.Errorf("Invalid DecodeParms")
			}
			decodeParams = dp
		}
	}

	// The EarlyChange indicates when to increase code length, as different
	// implementations use a different mechanisms. Essentially this chooses
	// which LZW implementation to use.
	// The default is 1 (one code early)
	obj, has := (*encDict)["EarlyChange"]
	if has {
		earlyChange, ok := obj.(*PdfObjectInteger)
		if !ok {
			common.Log.Debug("Error: EarlyChange specified but not numeric (%T)", obj)
			return nil, fmt.Errorf("Invalid EarlyChange")
		}
		if *earlyChange != 0 && *earlyChange != 1 {
			return nil, fmt.Errorf("Invalid EarlyChange value (not 0 or 1)")
		}

		encoder.EarlyChange = int(*earlyChange)
	} else {
		encoder.EarlyChange = 1 // default
	}

	if decodeParams == nil {
		// No decode parameters. Can safely return here if not set as the following options
		// are related to the decode Params.
		return encoder, nil
	}

	obj, has = (*decodeParams)["Predictor"]
	if has {
		predictor, ok := obj.(*PdfObjectInteger)
		if !ok {
			common.Log.Debug("Error: Predictor specified but not numeric (%T)", obj)
			return nil, fmt.Errorf("Invalid Predictor")
		}
		encoder.Predictor = int(*predictor)
	}

	// Bits per component.  Use default if not specified (8).
	obj, has = (*decodeParams)["BitsPerComponent"]
	if has {
		bpc, ok := obj.(*PdfObjectInteger)
		if !ok {
			common.Log.Debug("ERROR: Invalid BitsPerComponent")
			return nil, fmt.Errorf("Invalid BitsPerComponent")
		}
		encoder.BitsPerComponent = int(*bpc)
	}

	if encoder.Predictor > 1 {
		// Columns.
		encoder.Columns = 1
		obj, has = (*decodeParams)["Columns"]
		if has {
			columns, ok := obj.(*PdfObjectInteger)
			if !ok {
				return nil, fmt.Errorf("Predictor column invalid")
			}

			encoder.Columns = int(*columns)
		}

		// Colors.
		// Number of interleaved color components per sample (Default 1 if not specified)
		encoder.Colors = 1
		obj, has = (*decodeParams)["Colors"]
		if has {
			colors, ok := obj.(*PdfObjectInteger)
			if !ok {
				return nil, fmt.Errorf("Predictor colors not an integer")
			}
			encoder.Colors = int(*colors)
		}
	}

	common.Log.Trace("decode params: %s", decodeParams.String())
	return encoder, nil
}

func (this *LZWEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	var outBuf bytes.Buffer
	bufReader := bytes.NewReader(encoded)

	var r io.ReadCloser
	if this.EarlyChange == 1 {
		// LZW implementation with code length increases one code early (1).
		r = lzw1.NewReader(bufReader, lzw1.MSB, 8)
	} else {
		// 0: LZW implementation with postponed code length increases (0).
		r = lzw0.NewReader(bufReader, lzw0.MSB, 8)
	}
	defer r.Close()

	_, err := outBuf.ReadFrom(r)
	if err != nil {
		return nil, err
	}

	return outBuf.Bytes(), nil
}

func (this *LZWEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	// Revamp this support to handle TIFF predictor (2).
	// Also handle more filter bytes and check
	// BitsPerComponent.  Default value is 8, currently we are only
	// supporting that one.

	common.Log.Trace("LZW Decoding")
	common.Log.Trace("Predictor: %d", this.Predictor)

	outData, err := this.DecodeBytes(streamObj.Stream)
	if err != nil {
		return nil, err
	}

	common.Log.Trace(" IN: (%d) % x", len(streamObj.Stream), streamObj.Stream)
	common.Log.Trace("OUT: (%d) % x", len(outData), outData)

	if this.Predictor > 1 {
		if this.Predictor == 2 { // TIFF encoding: Needs some tests.
			common.Log.Trace("Tiff encoding")

			rowLength := int(this.Columns) * this.Colors
			rows := len(outData) / rowLength
			if len(outData)%rowLength != 0 {
				common.Log.Debug("ERROR: TIFF encoding: Invalid row length...")
				return nil, fmt.Errorf("Invalid row length (%d/%d)", len(outData), rowLength)
			}

			if rowLength%this.Colors != 0 {
				return nil, fmt.Errorf("Invalid row length (%d) for colors %d", rowLength, this.Colors)
			}
			common.Log.Trace("inp outData (%d): % x", len(outData), outData)

			pOutBuffer := bytes.NewBuffer(nil)

			// 0-255  -255 255 ; 0-255=-255;
			for i := 0; i < rows; i++ {
				rowData := outData[rowLength*i : rowLength*(i+1)]
				// Predicts the same as the sample to the left.
				// Interleaved by colors.
				for j := this.Colors; j < rowLength; j++ {
					rowData[j] = byte(int(rowData[j]+rowData[j-this.Colors]) % 256)
				}
				// GH: Appears that this is not working as expected...

				pOutBuffer.Write(rowData)
			}
			pOutData := pOutBuffer.Bytes()
			common.Log.Trace("POutData (%d): % x", len(pOutData), pOutData)
			return pOutData, nil
		} else if this.Predictor >= 10 && this.Predictor <= 15 {
			common.Log.Trace("PNG Encoding")
			// Columns represents the number of samples per row; Each sample can contain multiple color
			// components.
			rowLength := int(this.Columns*this.Colors + 1) // 1 byte to specify predictor algorithms per row.
			rows := len(outData) / rowLength
			if len(outData)%rowLength != 0 {
				return nil, fmt.Errorf("Invalid row length (%d/%d)", len(outData), rowLength)
			}

			pOutBuffer := bytes.NewBuffer(nil)

			common.Log.Trace("Predictor columns: %d", this.Columns)
			common.Log.Trace("Length: %d / %d = %d rows", len(outData), rowLength, rows)
			prevRowData := make([]byte, rowLength)
			for i := 0; i < rowLength; i++ {
				prevRowData[i] = 0
			}

			for i := 0; i < rows; i++ {
				rowData := outData[rowLength*i : rowLength*(i+1)]

				fb := rowData[0]
				switch fb {
				case 0:
					// No prediction. (No operation).
				case 1:
					// Sub: Predicts the same as the sample to the left.
					for j := 2; j < rowLength; j++ {
						rowData[j] = byte(int(rowData[j]+rowData[j-1]) % 256)
					}
				case 2:
					// Up: Predicts the same as the sample above
					for j := 1; j < rowLength; j++ {
						rowData[j] = byte(int(rowData[j]+prevRowData[j]) % 256)
					}
				default:
					common.Log.Debug("ERROR: Invalid filter byte (%d)", fb)
					return nil, fmt.Errorf("Invalid filter byte (%d)", fb)
				}

				for i := 0; i < rowLength; i++ {
					prevRowData[i] = rowData[i]
				}
				pOutBuffer.Write(rowData[1:])
			}
			pOutData := pOutBuffer.Bytes()
			return pOutData, nil
		} else {
			common.Log.Debug("ERROR: Unsupported predictor (%d)", this.Predictor)
			return nil, fmt.Errorf("Unsupported predictor (%d)", this.Predictor)
		}
	}

	return outData, nil
}

// Support for encoding LZW.  Currently not supporting predictors (raw compressed data only).
// Only supports the Early change = 1 algorithm (compress/lzw) as the other implementation
// does not have a write method.
// TODO: Consider refactoring compress/lzw to allow both.
func (this *LZWEncoder) EncodeBytes(data []byte) ([]byte, error) {
	if this.Predictor != 1 {
		return nil, fmt.Errorf("LZW Predictor = 1 only supported yet")
	}

	if this.EarlyChange == 1 {
		return nil, fmt.Errorf("LZW Early Change = 0 only supported yet")
	}

	var b bytes.Buffer
	w := lzw0.NewWriter(&b, lzw0.MSB, 8)
	w.Write(data)
	w.Close()

	return b.Bytes(), nil
}

//
// DCT (JPG) encoding/decoding functionality for images.
type DCTEncoder struct {
	ColorComponents  int // 1 (gray), 3 (rgb), 4 (cmyk)
	BitsPerComponent int // 8 or 16 bit
	Width            int
	Height           int
	Quality          int
}

// Make a new DCT encoder with default parameters.
func NewDCTEncoder() *DCTEncoder {
	encoder := &DCTEncoder{}

	encoder.ColorComponents = 3
	encoder.BitsPerComponent = 8

	encoder.Quality = 75

	return encoder
}

func (this *DCTEncoder) GetFilterName() string {
	return StreamEncodingFilterNameDCT
}

func (this *DCTEncoder) MakeDecodeParams() PdfObject {
	// Does not have decode params.
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
// Has the Filter set.  Some other parameters are generated elsewhere.
func (this *DCTEncoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())

	return &dict
}

// Create a new DCT encoder/decoder from a stream object, getting all the encoding parameters
// from the stream object dictionary entry and the image data itself.
// TODO: Support if used with other filters [ASCII85Decode FlateDecode DCTDecode]...
// need to apply the other filters prior to this one...
func newDCTEncoderFromStream(streamObj *PdfObjectStream, multiEnc *MultiEncoder) (*DCTEncoder, error) {
	// Start with default settings.
	encoder := NewDCTEncoder()

	encDict := streamObj.PdfObjectDictionary
	if encDict == nil {
		// No encoding dictionary.
		return encoder, nil
	}

	// If using DCTDecode in combination with other filters, make sure to decode that first...
	encoded := streamObj.Stream
	if multiEnc != nil {
		e, err := multiEnc.DecodeBytes(encoded)
		if err != nil {
			return nil, err
		}
		encoded = e

	}

	bufReader := bytes.NewReader(encoded)

	cfg, err := jpeg.DecodeConfig(bufReader)
	//img, _, err := goimage.Decode(bufReader)
	if err != nil {
		common.Log.Debug("Error decoding file: %s", err)
		return nil, err
	}

	switch cfg.ColorModel {
	case gocolor.RGBAModel:
		encoder.BitsPerComponent = 8
		encoder.ColorComponents = 3 // alpha is not included in pdf.
	case gocolor.RGBA64Model:
		encoder.BitsPerComponent = 16
		encoder.ColorComponents = 3
	case gocolor.GrayModel:
		encoder.BitsPerComponent = 8
		encoder.ColorComponents = 1
	case gocolor.Gray16Model:
		encoder.BitsPerComponent = 16
		encoder.ColorComponents = 1
	case gocolor.CMYKModel:
		encoder.BitsPerComponent = 8
		encoder.ColorComponents = 4
	case gocolor.YCbCrModel:
		// YCbCr is not supported by PDF, but it could be a different colorspace
		// with 3 components.  Would be specified by the ColorSpace entry.
		encoder.BitsPerComponent = 8
		encoder.ColorComponents = 3
	default:
		return nil, errors.New("Unsupported color model")
	}
	encoder.Width = cfg.Width
	encoder.Height = cfg.Height
	common.Log.Trace("DCT Encoder: %+v", encoder)
	encoder.Quality = 75

	return encoder, nil
}

func (this *DCTEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	bufReader := bytes.NewReader(encoded)
	//img, _, err := goimage.Decode(bufReader)
	img, err := jpeg.Decode(bufReader)
	if err != nil {
		common.Log.Debug("Error decoding image: %s", err)
		return nil, err
	}
	bounds := img.Bounds()

	var decoded = make([]byte, bounds.Dx()*bounds.Dy()*this.ColorComponents*this.BitsPerComponent/8)
	index := 0

	for j := bounds.Min.Y; j < bounds.Max.Y; j++ {
		for i := bounds.Min.X; i < bounds.Max.X; i++ {
			color := img.At(i, j)

			// Gray scale.
			if this.ColorComponents == 1 {
				if this.BitsPerComponent == 16 {
					// Gray - 16 bit.
					val, ok := color.(gocolor.Gray16)
					if !ok {
						return nil, errors.New("Color type error")
					}
					decoded[index] = byte((val.Y >> 8) & 0xff)
					index++
					decoded[index] = byte(val.Y & 0xff)
					index++
				} else {
					// Gray - 8 bit.
					val, ok := color.(gocolor.Gray)
					if !ok {
						return nil, errors.New("Color type error")
					}
					decoded[index] = byte(val.Y & 0xff)
					index++
				}
			} else if this.ColorComponents == 3 {
				if this.BitsPerComponent == 16 {
					val, ok := color.(gocolor.RGBA64)
					if !ok {
						return nil, errors.New("Color type error")
					}
					decoded[index] = byte((val.R >> 8) & 0xff)
					index++
					decoded[index] = byte(val.R & 0xff)
					index++
					decoded[index] = byte((val.G >> 8) & 0xff)
					index++
					decoded[index] = byte(val.G & 0xff)
					index++
					decoded[index] = byte((val.B >> 8) & 0xff)
					index++
					decoded[index] = byte(val.B & 0xff)
					index++
				} else {
					// RGB - 8 bit.
					val, isRGB := color.(gocolor.RGBA)
					if isRGB {
						decoded[index] = val.R & 0xff
						index++
						decoded[index] = val.G & 0xff
						index++
						decoded[index] = val.B & 0xff
						index++
					} else {
						// Hack around YCbCr from go jpeg package.
						val, ok := color.(gocolor.YCbCr)
						if !ok {
							return nil, errors.New("Color type error")
						}
						r, g, b, _ := val.RGBA()
						// The fact that we cannot use the Y, Cb, Cr values directly,
						// indicates that either the jpeg package is converting the raw
						// data into YCbCr with some kind of mapping, or that the original
						// data is not in R,G,B...
						// XXX: This is not good as it means we end up with R, G, B... even
						// if the original colormap was different.  Unless calling the RGBA()
						// call exactly reverses the previous conversion to YCbCr (even if
						// real data is not rgb)... ?
						// TODO: Test more. Consider whether we need to implement our own jpeg filter.
						decoded[index] = byte(r >> 8) //byte(val.Y & 0xff)
						index++
						decoded[index] = byte(g >> 8) //val.Cb & 0xff)
						index++
						decoded[index] = byte(b >> 8) //val.Cr & 0xff)
						index++
					}
				}
			} else if this.ColorComponents == 4 {
				// CMYK - 8 bit.
				val, ok := color.(gocolor.CMYK)
				if !ok {
					return nil, errors.New("Color type error")
				}
				// TODO: Is the inversion not handled right in the JPEG package for APP14?
				// Should not need to invert here...
				decoded[index] = 255 - val.C&0xff
				index++
				decoded[index] = 255 - val.M&0xff
				index++
				decoded[index] = 255 - val.Y&0xff
				index++
				decoded[index] = 255 - val.K&0xff
				index++
			}
		}
	}

	return decoded, nil
}

func (this *DCTEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	return this.DecodeBytes(streamObj.Stream)
}

type DrawableImage interface {
	ColorModel() gocolor.Model
	Bounds() goimage.Rectangle
	At(x, y int) gocolor.Color
	Set(x, y int, c gocolor.Color)
}

func (this *DCTEncoder) EncodeBytes(data []byte) ([]byte, error) {
	bounds := goimage.Rect(0, 0, this.Width, this.Height)
	var img DrawableImage
	if this.ColorComponents == 1 {
		if this.BitsPerComponent == 16 {
			img = goimage.NewGray16(bounds)
		} else {
			img = goimage.NewGray(bounds)
		}
	} else if this.ColorComponents == 3 {
		if this.BitsPerComponent == 16 {
			img = goimage.NewRGBA64(bounds)
		} else {
			img = goimage.NewRGBA(bounds)
		}
	} else if this.ColorComponents == 4 {
		img = goimage.NewCMYK(bounds)
	} else {
		return nil, errors.New("Unsupported")
	}

	// Draw the data on the image..
	x := 0
	y := 0
	bytesPerColor := this.ColorComponents * this.BitsPerComponent / 8
	for i := 0; i+bytesPerColor-1 < len(data); i += bytesPerColor {
		var c gocolor.Color
		if this.ColorComponents == 1 {
			if this.BitsPerComponent == 16 {
				val := uint16(data[i])<<8 | uint16(data[i+1])
				c = gocolor.Gray16{val}
			} else {
				val := uint8(data[i] & 0xff)
				c = gocolor.Gray{val}
			}
		} else if this.ColorComponents == 3 {
			if this.BitsPerComponent == 16 {
				r := uint16(data[i])<<8 | uint16(data[i+1])
				g := uint16(data[i+2])<<8 | uint16(data[i+3])
				b := uint16(data[i+4])<<8 | uint16(data[i+5])
				c = gocolor.RGBA64{R: r, G: g, B: b, A: 0}
			} else {
				r := uint8(data[i] & 0xff)
				g := uint8(data[i+1] & 0xff)
				b := uint8(data[i+2] & 0xff)
				c = gocolor.RGBA{R: r, G: g, B: b, A: 0}
			}
		} else if this.ColorComponents == 4 {
			c1 := uint8(data[i] & 0xff)
			m1 := uint8(data[i+1] & 0xff)
			y1 := uint8(data[i+2] & 0xff)
			k1 := uint8(data[i+3] & 0xff)
			c = gocolor.CMYK{C: c1, M: m1, Y: y1, K: k1}
		}

		img.Set(x, y, c)
		x++
		if x == this.Width {
			x = 0
			y++
		}
	}

	// The quality is specified from 0-100 (with 100 being the best quality) in the DCT structure.
	// N.B. even 100 is lossy, as still is transformed, but as good as it gets for DCT.
	// This is not related to the DPI, but rather inherent transformation losses.

	opt := jpeg.Options{}
	opt.Quality = this.Quality

	var buf bytes.Buffer
	err := jpeg.Encode(&buf, img, &opt)
	if err != nil {
		return nil, err
	}

	return buf.Bytes(), nil
}

/////
// ASCII hex encoder/decoder.
type ASCIIHexEncoder struct {
}

// Make a new ASCII hex encoder.
func NewASCIIHexEncoder() *ASCIIHexEncoder {
	encoder := &ASCIIHexEncoder{}
	return encoder
}

func (this *ASCIIHexEncoder) GetFilterName() string {
	return StreamEncodingFilterNameASCIIHex
}

func (this *ASCIIHexEncoder) MakeDecodeParams() PdfObject {
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
func (this *ASCIIHexEncoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())
	return &dict
}

func (this *ASCIIHexEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	bufReader := bytes.NewReader(encoded)
	inb := []byte{}
	for {
		b, err := bufReader.ReadByte()
		if err != nil {
			return nil, err
		}
		if b == '>' {
			break
		}
		if IsWhiteSpace(b) {
			continue
		}
		if (b >= 'a' && b <= 'f') || (b >= 'A' && b <= 'F') || (b >= '0' && b <= '9') {
			inb = append(inb, b)
		} else {
			common.Log.Debug("ERROR: Invalid ascii hex character (%c)", b)
			return nil, fmt.Errorf("Invalid ascii hex character (%c)", b)
		}
	}
	if len(inb)%2 == 1 {
		inb = append(inb, '0')
	}
	common.Log.Trace("Inbound %s", inb)
	outb := make([]byte, hex.DecodedLen(len(inb)))
	_, err := hex.Decode(outb, inb)
	if err != nil {
		return nil, err
	}
	return outb, nil
}

// ASCII hex decoding.
func (this *ASCIIHexEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	return this.DecodeBytes(streamObj.Stream)
}

func (this *ASCIIHexEncoder) EncodeBytes(data []byte) ([]byte, error) {
	var encoded bytes.Buffer

	for _, b := range data {
		encoded.WriteString(fmt.Sprintf("%.2X ", b))
	}
	encoded.WriteByte('>')

	return encoded.Bytes(), nil
}

//
// ASCII85 encoder/decoder.
//
type ASCII85Encoder struct {
}

// Make a new ASCII85 encoder.
func NewASCII85Encoder() *ASCII85Encoder {
	encoder := &ASCII85Encoder{}
	return encoder
}

func (this *ASCII85Encoder) GetFilterName() string {
	return StreamEncodingFilterNameASCII85
}

func (this *ASCII85Encoder) MakeDecodeParams() PdfObject {
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
func (this *ASCII85Encoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())
	return &dict
}

// 5 ASCII characters -> 4 raw binary bytes
func (this *ASCII85Encoder) DecodeBytes(encoded []byte) ([]byte, error) {
	decoded := []byte{}

	common.Log.Trace("ASCII85 Decode")

	i := 0
	eod := false

	for i < len(encoded) && !eod {
		codes := [5]byte{0, 0, 0, 0, 0}
		spaces := 0 // offset due to whitespace.
		j := 0
		toWrite := 4
		for j < 5+spaces {
			if i+j == len(encoded) {
				break
			}
			code := encoded[i+j]
			if IsWhiteSpace(code) {
				// Skip whitespace.
				spaces++
				j++
				continue
			} else if code == '~' && i+j+1 < len(encoded) && encoded[i+j+1] == '>' {
				toWrite = (j - spaces) - 1
				if toWrite < 0 {
					toWrite = 0
				}
				// EOD marker.  Marks end of data.
				eod = true
				break
			} else if code >= '!' && code <= 'u' {
				// Valid code.
				code -= '!'
			} else if code == 'z' && j-spaces == 0 {
				// 'z' in beginning of the byte sequence means that all 5 codes are 0.
				// Already all 0 initialized, so can break here.
				toWrite = 4
				j++
				break
			} else {
				common.Log.Error("Failed decoding, invalid code")
				return nil, errors.New("Invalid code encountered")
			}

			codes[j-spaces] = code
			j++
		}
		i += j

		// Pad with 'u' 84 (unused ones)
		// Takes care of issues at ends for input data that is not a multiple of 4-bytes.
		for m := toWrite + 1; m < 5; m++ {
			codes[m] = 84
		}

		// Convert to a uint32 value.
		value := uint32(codes[0])*85*85*85*85 + uint32(codes[1])*85*85*85 + uint32(codes[2])*85*85 + uint32(codes[3])*85 + uint32(codes[4])

		// Convert to 4 bytes.
		decodedBytes := []byte{
			byte((value >> 24) & 0xff),
			byte((value >> 16) & 0xff),
			byte((value >> 8) & 0xff),
			byte(value & 0xff)}

		// This accounts for the end of data, where the original data length is not a multiple of 4.
		// In that case, 0 bytes are assumed but only
		decoded = append(decoded, decodedBytes[:toWrite]...)
	}

	common.Log.Trace("ASCII85, encoded: % X", encoded)
	common.Log.Trace("ASCII85, decoded: % X", decoded)

	return decoded, nil
}

// ASCII85 stream decoding.
func (this *ASCII85Encoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	return this.DecodeBytes(streamObj.Stream)
}

// Convert a base 256 number to a series of base 85 values (5 codes).
//  85^5 = 4437053125 > 256^4 = 4294967296
// So 5 base-85 numbers will always be enough to cover 4 base-256 numbers.
// The base 256 value is already converted to an uint32 value.
func (this *ASCII85Encoder) base256Tobase85(base256val uint32) [5]byte {
	base85 := [5]byte{0, 0, 0, 0, 0}
	remainder := base256val
	for i := 0; i < 5; i++ {
		divider := uint32(1)
		for j := 0; j < 4-i; j++ {
			divider *= 85
		}
		val := remainder / divider
		remainder = remainder % divider
		base85[i] = byte(val)
	}
	return base85
}

// Encode data into ASCII85 encoded format.
func (this *ASCII85Encoder) EncodeBytes(data []byte) ([]byte, error) {
	var encoded bytes.Buffer

	for i := 0; i < len(data); i += 4 {
		b1 := data[i]
		n := 1

		b2 := byte(0)
		if i+1 < len(data) {
			b2 = data[i+1]
			n++
		}

		b3 := byte(0)
		if i+2 < len(data) {
			b3 = data[i+2]
			n++
		}

		b4 := byte(0)
		if i+3 < len(data) {
			b4 = data[i+3]
			n++
		}

		// Convert to a uint32 number.
		base256 := (uint32(b1) << 24) | (uint32(b2) << 16) | (uint32(b3) << 8) | uint32(b4)
		if base256 == 0 {
			encoded.WriteByte('z')
		} else {
			base85vals := this.base256Tobase85(base256)
			for _, val := range base85vals[:n+1] {
				encoded.WriteByte(val + '!')
			}
		}
	}

	// EOD.
	encoded.WriteString("~>")
	return encoded.Bytes(), nil
}

//
// Raw encoder/decoder (no encoding, pass through)
//
type RawEncoder struct{}

func NewRawEncoder() *RawEncoder {
	return &RawEncoder{}
}

func (this *RawEncoder) GetFilterName() string {
	return "Raw (no encoding)"
}

func (this *RawEncoder) MakeDecodeParams() PdfObject {
	return nil
}

// Make a new instance of an encoding dictionary for a stream object.
func (this *RawEncoder) MakeStreamDict() *PdfObjectDictionary {
	return &PdfObjectDictionary{}
}

func (this *RawEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	return encoded, nil
}

func (this *RawEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	return streamObj.Stream, nil
}

func (this *RawEncoder) EncodeBytes(data []byte) ([]byte, error) {
	return data, nil
}

//
// Multi encoder: support serial encoding.
//
type MultiEncoder struct {
	// Encoders in the order that they are to be applied.
	encoders []StreamEncoder
}

func NewMultiEncoder() *MultiEncoder {
	encoder := MultiEncoder{}
	encoder.encoders = []StreamEncoder{}

	return &encoder
}

func newMultiEncoderFromStream(streamObj *PdfObjectStream) (*MultiEncoder, error) {
	mencoder := NewMultiEncoder()

	encDict := streamObj.PdfObjectDictionary
	if encDict == nil {
		// No encoding dictionary.
		return mencoder, nil
	}

	// Prepare the decode params array (one for each filter type)
	// Optional, not always present.
	var decodeParamsDict *PdfObjectDictionary
	decodeParamsArray := []PdfObject{}
	obj := (*encDict)["DecodeParms"]
	if obj != nil {
		// If it is a dictionary, assume it applies to all
		dict, isDict := obj.(*PdfObjectDictionary)
		if isDict {
			decodeParamsDict = dict
		}

		// If it is an array, assume there is one for each
		arr, isArray := obj.(*PdfObjectArray)
		if isArray {
			for _, dictObj := range *arr {
				if dict, is := dictObj.(*PdfObjectDictionary); is {
					decodeParamsArray = append(decodeParamsArray, dict)
				} else {
					decodeParamsArray = append(decodeParamsArray, nil)
				}
			}
		}
	}

	obj = (*encDict)["Filter"]
	if obj == nil {
		return nil, fmt.Errorf("Filter missing")
	}

	array, ok := obj.(*PdfObjectArray)
	if !ok {
		return nil, fmt.Errorf("Multi filter can only be made from array")
	}

	for idx, obj := range *array {
		name, ok := obj.(*PdfObjectName)
		if !ok {
			return nil, fmt.Errorf("Multi filter array element not a name")
		}

		var dp PdfObject

		// If decode params dict is set, use it.  Otherwise take from array..
		if decodeParamsDict != nil {
			dp = decodeParamsDict
		} else {
			// Only get the dp if provided.  Oftentimes there is no decode params dict
			// provided.
			if len(decodeParamsArray) > 0 {
				if idx >= len(decodeParamsArray) {
					return nil, fmt.Errorf("Missing elements in decode params array")
				}
				dp = decodeParamsArray[idx]
			}
		}

		var dParams *PdfObjectDictionary
		if dict, is := dp.(*PdfObjectDictionary); is {
			dParams = dict
		}

		common.Log.Trace("Next name: %s", *name)
		if *name == StreamEncodingFilterNameFlate {
			// XXX: need to separate out the DecodeParms..
			encoder, err := newFlateEncoderFromStream(streamObj, dParams)
			if err != nil {
				return nil, err
			}
			mencoder.AddEncoder(encoder)
		} else if *name == StreamEncodingFilterNameLZW {
			encoder, err := newLZWEncoderFromStream(streamObj, dParams)
			if err != nil {
				return nil, err
			}
			mencoder.AddEncoder(encoder)
		} else if *name == StreamEncodingFilterNameASCIIHex {
			encoder := NewASCIIHexEncoder()
			mencoder.AddEncoder(encoder)
		} else if *name == StreamEncodingFilterNameASCII85 {
			encoder := NewASCII85Encoder()
			mencoder.AddEncoder(encoder)
		} else if *name == StreamEncodingFilterNameDCT {
			encoder, err := newDCTEncoderFromStream(streamObj, mencoder)
			if err != nil {
				return nil, err
			}
			mencoder.AddEncoder(encoder)
			common.Log.Trace("Added DCT encoder...")
			common.Log.Trace("Multi encoder: %#v", mencoder)
		} else {
			common.Log.Error("Unsupported filter %s", *name)
			return nil, fmt.Errorf("Invalid filter in multi filter array")
		}
	}

	return mencoder, nil
}

func (this *MultiEncoder) GetFilterName() string {
	name := ""
	for idx, encoder := range this.encoders {
		name += encoder.GetFilterName()
		if idx < len(this.encoders)-1 {
			name += " "
		}
	}
	return name
}

func (this *MultiEncoder) MakeDecodeParams() PdfObject {
	if len(this.encoders) == 0 {
		return nil
	}

	if len(this.encoders) == 1 {
		return this.encoders[0].MakeDecodeParams()
	}

	array := PdfObjectArray{}
	for _, encoder := range this.encoders {
		decodeParams := encoder.MakeDecodeParams()
		if decodeParams == nil {
			array = append(array, MakeNull())
		} else {
			array = append(array, decodeParams)
		}
	}

	return &array
}

func (this *MultiEncoder) AddEncoder(encoder StreamEncoder) {
	this.encoders = append(this.encoders, encoder)
}

func (this *MultiEncoder) MakeStreamDict() *PdfObjectDictionary {
	dict := PdfObjectDictionary{}

	dict["Filter"] = MakeName(this.GetFilterName())

	// Pass all values from children, except Filter and DecodeParms.
	for _, encoder := range this.encoders {
		encDict := encoder.MakeStreamDict()
		for key, val := range *encDict {
			if key != "Filter" && key != "DecodeParms" {
				dict[key] = val
			}
		}
	}

	// Make the decode params array or dict.
	decodeParams := this.MakeDecodeParams()
	if decodeParams != nil {
		dict["DecodeParms"] = decodeParams
	}

	return &dict
}

func (this *MultiEncoder) DecodeBytes(encoded []byte) ([]byte, error) {
	decoded := encoded
	var err error
	// Apply in forward order.
	for _, encoder := range this.encoders {
		common.Log.Trace("Multi Encoder Decode: Applying Filter: %v %T", encoder, encoder)

		decoded, err = encoder.DecodeBytes(decoded)
		if err != nil {
			return nil, err
		}
	}

	return decoded, nil
}

func (this *MultiEncoder) DecodeStream(streamObj *PdfObjectStream) ([]byte, error) {
	return this.DecodeBytes(streamObj.Stream)
}

func (this *MultiEncoder) EncodeBytes(data []byte) ([]byte, error) {
	encoded := data
	var err error

	// Apply in inverse order.
	for i := len(this.encoders) - 1; i >= 0; i-- {
		encoder := this.encoders[i]
		encoded, err = encoder.EncodeBytes(encoded)
		if err != nil {
			return nil, err
		}
	}

	return encoded, nil
}