Black Lives Matter. Support the Equal Justice Initiative.

Source file src/image/jpeg/reader.go

Documentation: image/jpeg

     1  // Copyright 2009 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package jpeg implements a JPEG image decoder and encoder.
     6  //
     7  // JPEG is defined in ITU-T T.81: https://www.w3.org/Graphics/JPEG/itu-t81.pdf.
     8  package jpeg
     9  
    10  import (
    11  	"image"
    12  	"image/color"
    13  	"image/internal/imageutil"
    14  	"io"
    15  )
    16  
    17  // A FormatError reports that the input is not a valid JPEG.
    18  type FormatError string
    19  
    20  func (e FormatError) Error() string { return "invalid JPEG format: " + string(e) }
    21  
    22  // An UnsupportedError reports that the input uses a valid but unimplemented JPEG feature.
    23  type UnsupportedError string
    24  
    25  func (e UnsupportedError) Error() string { return "unsupported JPEG feature: " + string(e) }
    26  
    27  var errUnsupportedSubsamplingRatio = UnsupportedError("luma/chroma subsampling ratio")
    28  
    29  // Component specification, specified in section B.2.2.
    30  type component struct {
    31  	h  int   // Horizontal sampling factor.
    32  	v  int   // Vertical sampling factor.
    33  	c  uint8 // Component identifier.
    34  	tq uint8 // Quantization table destination selector.
    35  }
    36  
    37  const (
    38  	dcTable = 0
    39  	acTable = 1
    40  	maxTc   = 1
    41  	maxTh   = 3
    42  	maxTq   = 3
    43  
    44  	maxComponents = 4
    45  )
    46  
    47  const (
    48  	sof0Marker = 0xc0 // Start Of Frame (Baseline Sequential).
    49  	sof1Marker = 0xc1 // Start Of Frame (Extended Sequential).
    50  	sof2Marker = 0xc2 // Start Of Frame (Progressive).
    51  	dhtMarker  = 0xc4 // Define Huffman Table.
    52  	rst0Marker = 0xd0 // ReSTart (0).
    53  	rst7Marker = 0xd7 // ReSTart (7).
    54  	soiMarker  = 0xd8 // Start Of Image.
    55  	eoiMarker  = 0xd9 // End Of Image.
    56  	sosMarker  = 0xda // Start Of Scan.
    57  	dqtMarker  = 0xdb // Define Quantization Table.
    58  	driMarker  = 0xdd // Define Restart Interval.
    59  	comMarker  = 0xfe // COMment.
    60  	// "APPlication specific" markers aren't part of the JPEG spec per se,
    61  	// but in practice, their use is described at
    62  	// https://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/JPEG.html
    63  	app0Marker  = 0xe0
    64  	app14Marker = 0xee
    65  	app15Marker = 0xef
    66  )
    67  
    68  // See https://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/JPEG.html#Adobe
    69  const (
    70  	adobeTransformUnknown = 0
    71  	adobeTransformYCbCr   = 1
    72  	adobeTransformYCbCrK  = 2
    73  )
    74  
    75  // unzig maps from the zig-zag ordering to the natural ordering. For example,
    76  // unzig[3] is the column and row of the fourth element in zig-zag order. The
    77  // value is 16, which means first column (16%8 == 0) and third row (16/8 == 2).
    78  var unzig = [blockSize]int{
    79  	0, 1, 8, 16, 9, 2, 3, 10,
    80  	17, 24, 32, 25, 18, 11, 4, 5,
    81  	12, 19, 26, 33, 40, 48, 41, 34,
    82  	27, 20, 13, 6, 7, 14, 21, 28,
    83  	35, 42, 49, 56, 57, 50, 43, 36,
    84  	29, 22, 15, 23, 30, 37, 44, 51,
    85  	58, 59, 52, 45, 38, 31, 39, 46,
    86  	53, 60, 61, 54, 47, 55, 62, 63,
    87  }
    88  
    89  // Deprecated: Reader is not used by the image/jpeg package and should
    90  // not be used by others. It is kept for compatibility.
    91  type Reader interface {
    92  	io.ByteReader
    93  	io.Reader
    94  }
    95  
    96  // bits holds the unprocessed bits that have been taken from the byte-stream.
    97  // The n least significant bits of a form the unread bits, to be read in MSB to
    98  // LSB order.
    99  type bits struct {
   100  	a uint32 // accumulator.
   101  	m uint32 // mask. m==1<<(n-1) when n>0, with m==0 when n==0.
   102  	n int32  // the number of unread bits in a.
   103  }
   104  
   105  type decoder struct {
   106  	r    io.Reader
   107  	bits bits
   108  	// bytes is a byte buffer, similar to a bufio.Reader, except that it
   109  	// has to be able to unread more than 1 byte, due to byte stuffing.
   110  	// Byte stuffing is specified in section F.1.2.3.
   111  	bytes struct {
   112  		// buf[i:j] are the buffered bytes read from the underlying
   113  		// io.Reader that haven't yet been passed further on.
   114  		buf  [4096]byte
   115  		i, j int
   116  		// nUnreadable is the number of bytes to back up i after
   117  		// overshooting. It can be 0, 1 or 2.
   118  		nUnreadable int
   119  	}
   120  	width, height int
   121  
   122  	img1        *image.Gray
   123  	img3        *image.YCbCr
   124  	blackPix    []byte
   125  	blackStride int
   126  
   127  	ri    int // Restart Interval.
   128  	nComp int
   129  
   130  	// As per section 4.5, there are four modes of operation (selected by the
   131  	// SOF? markers): sequential DCT, progressive DCT, lossless and
   132  	// hierarchical, although this implementation does not support the latter
   133  	// two non-DCT modes. Sequential DCT is further split into baseline and
   134  	// extended, as per section 4.11.
   135  	baseline    bool
   136  	progressive bool
   137  
   138  	jfif                bool
   139  	adobeTransformValid bool
   140  	adobeTransform      uint8
   141  	eobRun              uint16 // End-of-Band run, specified in section G.1.2.2.
   142  
   143  	comp       [maxComponents]component
   144  	progCoeffs [maxComponents][]block // Saved state between progressive-mode scans.
   145  	huff       [maxTc + 1][maxTh + 1]huffman
   146  	quant      [maxTq + 1]block // Quantization tables, in zig-zag order.
   147  	tmp        [2 * blockSize]byte
   148  }
   149  
   150  // fill fills up the d.bytes.buf buffer from the underlying io.Reader. It
   151  // should only be called when there are no unread bytes in d.bytes.
   152  func (d *decoder) fill() error {
   153  	if d.bytes.i != d.bytes.j {
   154  		panic("jpeg: fill called when unread bytes exist")
   155  	}
   156  	// Move the last 2 bytes to the start of the buffer, in case we need
   157  	// to call unreadByteStuffedByte.
   158  	if d.bytes.j > 2 {
   159  		d.bytes.buf[0] = d.bytes.buf[d.bytes.j-2]
   160  		d.bytes.buf[1] = d.bytes.buf[d.bytes.j-1]
   161  		d.bytes.i, d.bytes.j = 2, 2
   162  	}
   163  	// Fill in the rest of the buffer.
   164  	n, err := d.r.Read(d.bytes.buf[d.bytes.j:])
   165  	d.bytes.j += n
   166  	if n > 0 {
   167  		err = nil
   168  	}
   169  	return err
   170  }
   171  
   172  // unreadByteStuffedByte undoes the most recent readByteStuffedByte call,
   173  // giving a byte of data back from d.bits to d.bytes. The Huffman look-up table
   174  // requires at least 8 bits for look-up, which means that Huffman decoding can
   175  // sometimes overshoot and read one or two too many bytes. Two-byte overshoot
   176  // can happen when expecting to read a 0xff 0x00 byte-stuffed byte.
   177  func (d *decoder) unreadByteStuffedByte() {
   178  	d.bytes.i -= d.bytes.nUnreadable
   179  	d.bytes.nUnreadable = 0
   180  	if d.bits.n >= 8 {
   181  		d.bits.a >>= 8
   182  		d.bits.n -= 8
   183  		d.bits.m >>= 8
   184  	}
   185  }
   186  
   187  // readByte returns the next byte, whether buffered or not buffered. It does
   188  // not care about byte stuffing.
   189  func (d *decoder) readByte() (x byte, err error) {
   190  	for d.bytes.i == d.bytes.j {
   191  		if err = d.fill(); err != nil {
   192  			return 0, err
   193  		}
   194  	}
   195  	x = d.bytes.buf[d.bytes.i]
   196  	d.bytes.i++
   197  	d.bytes.nUnreadable = 0
   198  	return x, nil
   199  }
   200  
   201  // errMissingFF00 means that readByteStuffedByte encountered an 0xff byte (a
   202  // marker byte) that wasn't the expected byte-stuffed sequence 0xff, 0x00.
   203  var errMissingFF00 = FormatError("missing 0xff00 sequence")
   204  
   205  // readByteStuffedByte is like readByte but is for byte-stuffed Huffman data.
   206  func (d *decoder) readByteStuffedByte() (x byte, err error) {
   207  	// Take the fast path if d.bytes.buf contains at least two bytes.
   208  	if d.bytes.i+2 <= d.bytes.j {
   209  		x = d.bytes.buf[d.bytes.i]
   210  		d.bytes.i++
   211  		d.bytes.nUnreadable = 1
   212  		if x != 0xff {
   213  			return x, err
   214  		}
   215  		if d.bytes.buf[d.bytes.i] != 0x00 {
   216  			return 0, errMissingFF00
   217  		}
   218  		d.bytes.i++
   219  		d.bytes.nUnreadable = 2
   220  		return 0xff, nil
   221  	}
   222  
   223  	d.bytes.nUnreadable = 0
   224  
   225  	x, err = d.readByte()
   226  	if err != nil {
   227  		return 0, err
   228  	}
   229  	d.bytes.nUnreadable = 1
   230  	if x != 0xff {
   231  		return x, nil
   232  	}
   233  
   234  	x, err = d.readByte()
   235  	if err != nil {
   236  		return 0, err
   237  	}
   238  	d.bytes.nUnreadable = 2
   239  	if x != 0x00 {
   240  		return 0, errMissingFF00
   241  	}
   242  	return 0xff, nil
   243  }
   244  
   245  // readFull reads exactly len(p) bytes into p. It does not care about byte
   246  // stuffing.
   247  func (d *decoder) readFull(p []byte) error {
   248  	// Unread the overshot bytes, if any.
   249  	if d.bytes.nUnreadable != 0 {
   250  		if d.bits.n >= 8 {
   251  			d.unreadByteStuffedByte()
   252  		}
   253  		d.bytes.nUnreadable = 0
   254  	}
   255  
   256  	for {
   257  		n := copy(p, d.bytes.buf[d.bytes.i:d.bytes.j])
   258  		p = p[n:]
   259  		d.bytes.i += n
   260  		if len(p) == 0 {
   261  			break
   262  		}
   263  		if err := d.fill(); err != nil {
   264  			if err == io.EOF {
   265  				err = io.ErrUnexpectedEOF
   266  			}
   267  			return err
   268  		}
   269  	}
   270  	return nil
   271  }
   272  
   273  // ignore ignores the next n bytes.
   274  func (d *decoder) ignore(n int) error {
   275  	// Unread the overshot bytes, if any.
   276  	if d.bytes.nUnreadable != 0 {
   277  		if d.bits.n >= 8 {
   278  			d.unreadByteStuffedByte()
   279  		}
   280  		d.bytes.nUnreadable = 0
   281  	}
   282  
   283  	for {
   284  		m := d.bytes.j - d.bytes.i
   285  		if m > n {
   286  			m = n
   287  		}
   288  		d.bytes.i += m
   289  		n -= m
   290  		if n == 0 {
   291  			break
   292  		}
   293  		if err := d.fill(); err != nil {
   294  			if err == io.EOF {
   295  				err = io.ErrUnexpectedEOF
   296  			}
   297  			return err
   298  		}
   299  	}
   300  	return nil
   301  }
   302  
   303  // Specified in section B.2.2.
   304  func (d *decoder) processSOF(n int) error {
   305  	if d.nComp != 0 {
   306  		return FormatError("multiple SOF markers")
   307  	}
   308  	switch n {
   309  	case 6 + 3*1: // Grayscale image.
   310  		d.nComp = 1
   311  	case 6 + 3*3: // YCbCr or RGB image.
   312  		d.nComp = 3
   313  	case 6 + 3*4: // YCbCrK or CMYK image.
   314  		d.nComp = 4
   315  	default:
   316  		return UnsupportedError("number of components")
   317  	}
   318  	if err := d.readFull(d.tmp[:n]); err != nil {
   319  		return err
   320  	}
   321  	// We only support 8-bit precision.
   322  	if d.tmp[0] != 8 {
   323  		return UnsupportedError("precision")
   324  	}
   325  	d.height = int(d.tmp[1])<<8 + int(d.tmp[2])
   326  	d.width = int(d.tmp[3])<<8 + int(d.tmp[4])
   327  	if int(d.tmp[5]) != d.nComp {
   328  		return FormatError("SOF has wrong length")
   329  	}
   330  
   331  	for i := 0; i < d.nComp; i++ {
   332  		d.comp[i].c = d.tmp[6+3*i]
   333  		// Section B.2.2 states that "the value of C_i shall be different from
   334  		// the values of C_1 through C_(i-1)".
   335  		for j := 0; j < i; j++ {
   336  			if d.comp[i].c == d.comp[j].c {
   337  				return FormatError("repeated component identifier")
   338  			}
   339  		}
   340  
   341  		d.comp[i].tq = d.tmp[8+3*i]
   342  		if d.comp[i].tq > maxTq {
   343  			return FormatError("bad Tq value")
   344  		}
   345  
   346  		hv := d.tmp[7+3*i]
   347  		h, v := int(hv>>4), int(hv&0x0f)
   348  		if h < 1 || 4 < h || v < 1 || 4 < v {
   349  			return FormatError("luma/chroma subsampling ratio")
   350  		}
   351  		if h == 3 || v == 3 {
   352  			return errUnsupportedSubsamplingRatio
   353  		}
   354  		switch d.nComp {
   355  		case 1:
   356  			// If a JPEG image has only one component, section A.2 says "this data
   357  			// is non-interleaved by definition" and section A.2.2 says "[in this
   358  			// case...] the order of data units within a scan shall be left-to-right
   359  			// and top-to-bottom... regardless of the values of H_1 and V_1". Section
   360  			// 4.8.2 also says "[for non-interleaved data], the MCU is defined to be
   361  			// one data unit". Similarly, section A.1.1 explains that it is the ratio
   362  			// of H_i to max_j(H_j) that matters, and similarly for V. For grayscale
   363  			// images, H_1 is the maximum H_j for all components j, so that ratio is
   364  			// always 1. The component's (h, v) is effectively always (1, 1): even if
   365  			// the nominal (h, v) is (2, 1), a 20x5 image is encoded in three 8x8
   366  			// MCUs, not two 16x8 MCUs.
   367  			h, v = 1, 1
   368  
   369  		case 3:
   370  			// For YCbCr images, we only support 4:4:4, 4:4:0, 4:2:2, 4:2:0,
   371  			// 4:1:1 or 4:1:0 chroma subsampling ratios. This implies that the
   372  			// (h, v) values for the Y component are either (1, 1), (1, 2),
   373  			// (2, 1), (2, 2), (4, 1) or (4, 2), and the Y component's values
   374  			// must be a multiple of the Cb and Cr component's values. We also
   375  			// assume that the two chroma components have the same subsampling
   376  			// ratio.
   377  			switch i {
   378  			case 0: // Y.
   379  				// We have already verified, above, that h and v are both
   380  				// either 1, 2 or 4, so invalid (h, v) combinations are those
   381  				// with v == 4.
   382  				if v == 4 {
   383  					return errUnsupportedSubsamplingRatio
   384  				}
   385  			case 1: // Cb.
   386  				if d.comp[0].h%h != 0 || d.comp[0].v%v != 0 {
   387  					return errUnsupportedSubsamplingRatio
   388  				}
   389  			case 2: // Cr.
   390  				if d.comp[1].h != h || d.comp[1].v != v {
   391  					return errUnsupportedSubsamplingRatio
   392  				}
   393  			}
   394  
   395  		case 4:
   396  			// For 4-component images (either CMYK or YCbCrK), we only support two
   397  			// hv vectors: [0x11 0x11 0x11 0x11] and [0x22 0x11 0x11 0x22].
   398  			// Theoretically, 4-component JPEG images could mix and match hv values
   399  			// but in practice, those two combinations are the only ones in use,
   400  			// and it simplifies the applyBlack code below if we can assume that:
   401  			//	- for CMYK, the C and K channels have full samples, and if the M
   402  			//	  and Y channels subsample, they subsample both horizontally and
   403  			//	  vertically.
   404  			//	- for YCbCrK, the Y and K channels have full samples.
   405  			switch i {
   406  			case 0:
   407  				if hv != 0x11 && hv != 0x22 {
   408  					return errUnsupportedSubsamplingRatio
   409  				}
   410  			case 1, 2:
   411  				if hv != 0x11 {
   412  					return errUnsupportedSubsamplingRatio
   413  				}
   414  			case 3:
   415  				if d.comp[0].h != h || d.comp[0].v != v {
   416  					return errUnsupportedSubsamplingRatio
   417  				}
   418  			}
   419  		}
   420  
   421  		d.comp[i].h = h
   422  		d.comp[i].v = v
   423  	}
   424  	return nil
   425  }
   426  
   427  // Specified in section B.2.4.1.
   428  func (d *decoder) processDQT(n int) error {
   429  loop:
   430  	for n > 0 {
   431  		n--
   432  		x, err := d.readByte()
   433  		if err != nil {
   434  			return err
   435  		}
   436  		tq := x & 0x0f
   437  		if tq > maxTq {
   438  			return FormatError("bad Tq value")
   439  		}
   440  		switch x >> 4 {
   441  		default:
   442  			return FormatError("bad Pq value")
   443  		case 0:
   444  			if n < blockSize {
   445  				break loop
   446  			}
   447  			n -= blockSize
   448  			if err := d.readFull(d.tmp[:blockSize]); err != nil {
   449  				return err
   450  			}
   451  			for i := range d.quant[tq] {
   452  				d.quant[tq][i] = int32(d.tmp[i])
   453  			}
   454  		case 1:
   455  			if n < 2*blockSize {
   456  				break loop
   457  			}
   458  			n -= 2 * blockSize
   459  			if err := d.readFull(d.tmp[:2*blockSize]); err != nil {
   460  				return err
   461  			}
   462  			for i := range d.quant[tq] {
   463  				d.quant[tq][i] = int32(d.tmp[2*i])<<8 | int32(d.tmp[2*i+1])
   464  			}
   465  		}
   466  	}
   467  	if n != 0 {
   468  		return FormatError("DQT has wrong length")
   469  	}
   470  	return nil
   471  }
   472  
   473  // Specified in section B.2.4.4.
   474  func (d *decoder) processDRI(n int) error {
   475  	if n != 2 {
   476  		return FormatError("DRI has wrong length")
   477  	}
   478  	if err := d.readFull(d.tmp[:2]); err != nil {
   479  		return err
   480  	}
   481  	d.ri = int(d.tmp[0])<<8 + int(d.tmp[1])
   482  	return nil
   483  }
   484  
   485  func (d *decoder) processApp0Marker(n int) error {
   486  	if n < 5 {
   487  		return d.ignore(n)
   488  	}
   489  	if err := d.readFull(d.tmp[:5]); err != nil {
   490  		return err
   491  	}
   492  	n -= 5
   493  
   494  	d.jfif = d.tmp[0] == 'J' && d.tmp[1] == 'F' && d.tmp[2] == 'I' && d.tmp[3] == 'F' && d.tmp[4] == '\x00'
   495  
   496  	if n > 0 {
   497  		return d.ignore(n)
   498  	}
   499  	return nil
   500  }
   501  
   502  func (d *decoder) processApp14Marker(n int) error {
   503  	if n < 12 {
   504  		return d.ignore(n)
   505  	}
   506  	if err := d.readFull(d.tmp[:12]); err != nil {
   507  		return err
   508  	}
   509  	n -= 12
   510  
   511  	if d.tmp[0] == 'A' && d.tmp[1] == 'd' && d.tmp[2] == 'o' && d.tmp[3] == 'b' && d.tmp[4] == 'e' {
   512  		d.adobeTransformValid = true
   513  		d.adobeTransform = d.tmp[11]
   514  	}
   515  
   516  	if n > 0 {
   517  		return d.ignore(n)
   518  	}
   519  	return nil
   520  }
   521  
   522  // decode reads a JPEG image from r and returns it as an image.Image.
   523  func (d *decoder) decode(r io.Reader, configOnly bool) (image.Image, error) {
   524  	d.r = r
   525  
   526  	// Check for the Start Of Image marker.
   527  	if err := d.readFull(d.tmp[:2]); err != nil {
   528  		return nil, err
   529  	}
   530  	if d.tmp[0] != 0xff || d.tmp[1] != soiMarker {
   531  		return nil, FormatError("missing SOI marker")
   532  	}
   533  
   534  	// Process the remaining segments until the End Of Image marker.
   535  	for {
   536  		err := d.readFull(d.tmp[:2])
   537  		if err != nil {
   538  			return nil, err
   539  		}
   540  		for d.tmp[0] != 0xff {
   541  			// Strictly speaking, this is a format error. However, libjpeg is
   542  			// liberal in what it accepts. As of version 9, next_marker in
   543  			// jdmarker.c treats this as a warning (JWRN_EXTRANEOUS_DATA) and
   544  			// continues to decode the stream. Even before next_marker sees
   545  			// extraneous data, jpeg_fill_bit_buffer in jdhuff.c reads as many
   546  			// bytes as it can, possibly past the end of a scan's data. It
   547  			// effectively puts back any markers that it overscanned (e.g. an
   548  			// "\xff\xd9" EOI marker), but it does not put back non-marker data,
   549  			// and thus it can silently ignore a small number of extraneous
   550  			// non-marker bytes before next_marker has a chance to see them (and
   551  			// print a warning).
   552  			//
   553  			// We are therefore also liberal in what we accept. Extraneous data
   554  			// is silently ignored.
   555  			//
   556  			// This is similar to, but not exactly the same as, the restart
   557  			// mechanism within a scan (the RST[0-7] markers).
   558  			//
   559  			// Note that extraneous 0xff bytes in e.g. SOS data are escaped as
   560  			// "\xff\x00", and so are detected a little further down below.
   561  			d.tmp[0] = d.tmp[1]
   562  			d.tmp[1], err = d.readByte()
   563  			if err != nil {
   564  				return nil, err
   565  			}
   566  		}
   567  		marker := d.tmp[1]
   568  		if marker == 0 {
   569  			// Treat "\xff\x00" as extraneous data.
   570  			continue
   571  		}
   572  		for marker == 0xff {
   573  			// Section B.1.1.2 says, "Any marker may optionally be preceded by any
   574  			// number of fill bytes, which are bytes assigned code X'FF'".
   575  			marker, err = d.readByte()
   576  			if err != nil {
   577  				return nil, err
   578  			}
   579  		}
   580  		if marker == eoiMarker { // End Of Image.
   581  			break
   582  		}
   583  		if rst0Marker <= marker && marker <= rst7Marker {
   584  			// Figures B.2 and B.16 of the specification suggest that restart markers should
   585  			// only occur between Entropy Coded Segments and not after the final ECS.
   586  			// However, some encoders may generate incorrect JPEGs with a final restart
   587  			// marker. That restart marker will be seen here instead of inside the processSOS
   588  			// method, and is ignored as a harmless error. Restart markers have no extra data,
   589  			// so we check for this before we read the 16-bit length of the segment.
   590  			continue
   591  		}
   592  
   593  		// Read the 16-bit length of the segment. The value includes the 2 bytes for the
   594  		// length itself, so we subtract 2 to get the number of remaining bytes.
   595  		if err = d.readFull(d.tmp[:2]); err != nil {
   596  			return nil, err
   597  		}
   598  		n := int(d.tmp[0])<<8 + int(d.tmp[1]) - 2
   599  		if n < 0 {
   600  			return nil, FormatError("short segment length")
   601  		}
   602  
   603  		switch marker {
   604  		case sof0Marker, sof1Marker, sof2Marker:
   605  			d.baseline = marker == sof0Marker
   606  			d.progressive = marker == sof2Marker
   607  			err = d.processSOF(n)
   608  			if configOnly && d.jfif {
   609  				return nil, err
   610  			}
   611  		case dhtMarker:
   612  			if configOnly {
   613  				err = d.ignore(n)
   614  			} else {
   615  				err = d.processDHT(n)
   616  			}
   617  		case dqtMarker:
   618  			if configOnly {
   619  				err = d.ignore(n)
   620  			} else {
   621  				err = d.processDQT(n)
   622  			}
   623  		case sosMarker:
   624  			if configOnly {
   625  				return nil, nil
   626  			}
   627  			err = d.processSOS(n)
   628  		case driMarker:
   629  			if configOnly {
   630  				err = d.ignore(n)
   631  			} else {
   632  				err = d.processDRI(n)
   633  			}
   634  		case app0Marker:
   635  			err = d.processApp0Marker(n)
   636  		case app14Marker:
   637  			err = d.processApp14Marker(n)
   638  		default:
   639  			if app0Marker <= marker && marker <= app15Marker || marker == comMarker {
   640  				err = d.ignore(n)
   641  			} else if marker < 0xc0 { // See Table B.1 "Marker code assignments".
   642  				err = FormatError("unknown marker")
   643  			} else {
   644  				err = UnsupportedError("unknown marker")
   645  			}
   646  		}
   647  		if err != nil {
   648  			return nil, err
   649  		}
   650  	}
   651  
   652  	if d.progressive {
   653  		if err := d.reconstructProgressiveImage(); err != nil {
   654  			return nil, err
   655  		}
   656  	}
   657  	if d.img1 != nil {
   658  		return d.img1, nil
   659  	}
   660  	if d.img3 != nil {
   661  		if d.blackPix != nil {
   662  			return d.applyBlack()
   663  		} else if d.isRGB() {
   664  			return d.convertToRGB()
   665  		}
   666  		return d.img3, nil
   667  	}
   668  	return nil, FormatError("missing SOS marker")
   669  }
   670  
   671  // applyBlack combines d.img3 and d.blackPix into a CMYK image. The formula
   672  // used depends on whether the JPEG image is stored as CMYK or YCbCrK,
   673  // indicated by the APP14 (Adobe) metadata.
   674  //
   675  // Adobe CMYK JPEG images are inverted, where 255 means no ink instead of full
   676  // ink, so we apply "v = 255 - v" at various points. Note that a double
   677  // inversion is a no-op, so inversions might be implicit in the code below.
   678  func (d *decoder) applyBlack() (image.Image, error) {
   679  	if !d.adobeTransformValid {
   680  		return nil, UnsupportedError("unknown color model: 4-component JPEG doesn't have Adobe APP14 metadata")
   681  	}
   682  
   683  	// If the 4-component JPEG image isn't explicitly marked as "Unknown (RGB
   684  	// or CMYK)" as per
   685  	// https://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/JPEG.html#Adobe
   686  	// we assume that it is YCbCrK. This matches libjpeg's jdapimin.c.
   687  	if d.adobeTransform != adobeTransformUnknown {
   688  		// Convert the YCbCr part of the YCbCrK to RGB, invert the RGB to get
   689  		// CMY, and patch in the original K. The RGB to CMY inversion cancels
   690  		// out the 'Adobe inversion' described in the applyBlack doc comment
   691  		// above, so in practice, only the fourth channel (black) is inverted.
   692  		bounds := d.img3.Bounds()
   693  		img := image.NewRGBA(bounds)
   694  		imageutil.DrawYCbCr(img, bounds, d.img3, bounds.Min)
   695  		for iBase, y := 0, bounds.Min.Y; y < bounds.Max.Y; iBase, y = iBase+img.Stride, y+1 {
   696  			for i, x := iBase+3, bounds.Min.X; x < bounds.Max.X; i, x = i+4, x+1 {
   697  				img.Pix[i] = 255 - d.blackPix[(y-bounds.Min.Y)*d.blackStride+(x-bounds.Min.X)]
   698  			}
   699  		}
   700  		return &image.CMYK{
   701  			Pix:    img.Pix,
   702  			Stride: img.Stride,
   703  			Rect:   img.Rect,
   704  		}, nil
   705  	}
   706  
   707  	// The first three channels (cyan, magenta, yellow) of the CMYK
   708  	// were decoded into d.img3, but each channel was decoded into a separate
   709  	// []byte slice, and some channels may be subsampled. We interleave the
   710  	// separate channels into an image.CMYK's single []byte slice containing 4
   711  	// contiguous bytes per pixel.
   712  	bounds := d.img3.Bounds()
   713  	img := image.NewCMYK(bounds)
   714  
   715  	translations := [4]struct {
   716  		src    []byte
   717  		stride int
   718  	}{
   719  		{d.img3.Y, d.img3.YStride},
   720  		{d.img3.Cb, d.img3.CStride},
   721  		{d.img3.Cr, d.img3.CStride},
   722  		{d.blackPix, d.blackStride},
   723  	}
   724  	for t, translation := range translations {
   725  		subsample := d.comp[t].h != d.comp[0].h || d.comp[t].v != d.comp[0].v
   726  		for iBase, y := 0, bounds.Min.Y; y < bounds.Max.Y; iBase, y = iBase+img.Stride, y+1 {
   727  			sy := y - bounds.Min.Y
   728  			if subsample {
   729  				sy /= 2
   730  			}
   731  			for i, x := iBase+t, bounds.Min.X; x < bounds.Max.X; i, x = i+4, x+1 {
   732  				sx := x - bounds.Min.X
   733  				if subsample {
   734  					sx /= 2
   735  				}
   736  				img.Pix[i] = 255 - translation.src[sy*translation.stride+sx]
   737  			}
   738  		}
   739  	}
   740  	return img, nil
   741  }
   742  
   743  func (d *decoder) isRGB() bool {
   744  	if d.jfif {
   745  		return false
   746  	}
   747  	if d.adobeTransformValid && d.adobeTransform == adobeTransformUnknown {
   748  		// https://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/JPEG.html#Adobe
   749  		// says that 0 means Unknown (and in practice RGB) and 1 means YCbCr.
   750  		return true
   751  	}
   752  	return d.comp[0].c == 'R' && d.comp[1].c == 'G' && d.comp[2].c == 'B'
   753  }
   754  
   755  func (d *decoder) convertToRGB() (image.Image, error) {
   756  	cScale := d.comp[0].h / d.comp[1].h
   757  	bounds := d.img3.Bounds()
   758  	img := image.NewRGBA(bounds)
   759  	for y := bounds.Min.Y; y < bounds.Max.Y; y++ {
   760  		po := img.PixOffset(bounds.Min.X, y)
   761  		yo := d.img3.YOffset(bounds.Min.X, y)
   762  		co := d.img3.COffset(bounds.Min.X, y)
   763  		for i, iMax := 0, bounds.Max.X-bounds.Min.X; i < iMax; i++ {
   764  			img.Pix[po+4*i+0] = d.img3.Y[yo+i]
   765  			img.Pix[po+4*i+1] = d.img3.Cb[co+i/cScale]
   766  			img.Pix[po+4*i+2] = d.img3.Cr[co+i/cScale]
   767  			img.Pix[po+4*i+3] = 255
   768  		}
   769  	}
   770  	return img, nil
   771  }
   772  
   773  // Decode reads a JPEG image from r and returns it as an image.Image.
   774  func Decode(r io.Reader) (image.Image, error) {
   775  	var d decoder
   776  	return d.decode(r, false)
   777  }
   778  
   779  // DecodeConfig returns the color model and dimensions of a JPEG image without
   780  // decoding the entire image.
   781  func DecodeConfig(r io.Reader) (image.Config, error) {
   782  	var d decoder
   783  	if _, err := d.decode(r, true); err != nil {
   784  		return image.Config{}, err
   785  	}
   786  	switch d.nComp {
   787  	case 1:
   788  		return image.Config{
   789  			ColorModel: color.GrayModel,
   790  			Width:      d.width,
   791  			Height:     d.height,
   792  		}, nil
   793  	case 3:
   794  		cm := color.YCbCrModel
   795  		if d.isRGB() {
   796  			cm = color.RGBAModel
   797  		}
   798  		return image.Config{
   799  			ColorModel: cm,
   800  			Width:      d.width,
   801  			Height:     d.height,
   802  		}, nil
   803  	case 4:
   804  		return image.Config{
   805  			ColorModel: color.CMYKModel,
   806  			Width:      d.width,
   807  			Height:     d.height,
   808  		}, nil
   809  	}
   810  	return image.Config{}, FormatError("missing SOF marker")
   811  }
   812  
   813  func init() {
   814  	image.RegisterFormat("jpeg", "\xff\xd8", Decode, DecodeConfig)
   815  }
   816  

View as plain text