...

Source file src/image/jpeg/scan.go

     1	// Copyright 2012 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
     6	
     7	import (
     8		"image"
     9	)
    10	
    11	// makeImg allocates and initializes the destination image.
    12	func (d *decoder) makeImg(mxx, myy int) {
    13		if d.nComp == 1 {
    14			m := image.NewGray(image.Rect(0, 0, 8*mxx, 8*myy))
    15			d.img1 = m.SubImage(image.Rect(0, 0, d.width, d.height)).(*image.Gray)
    16			return
    17		}
    18	
    19		h0 := d.comp[0].h
    20		v0 := d.comp[0].v
    21		hRatio := h0 / d.comp[1].h
    22		vRatio := v0 / d.comp[1].v
    23		var subsampleRatio image.YCbCrSubsampleRatio
    24		switch hRatio<<4 | vRatio {
    25		case 0x11:
    26			subsampleRatio = image.YCbCrSubsampleRatio444
    27		case 0x12:
    28			subsampleRatio = image.YCbCrSubsampleRatio440
    29		case 0x21:
    30			subsampleRatio = image.YCbCrSubsampleRatio422
    31		case 0x22:
    32			subsampleRatio = image.YCbCrSubsampleRatio420
    33		case 0x41:
    34			subsampleRatio = image.YCbCrSubsampleRatio411
    35		case 0x42:
    36			subsampleRatio = image.YCbCrSubsampleRatio410
    37		default:
    38			panic("unreachable")
    39		}
    40		m := image.NewYCbCr(image.Rect(0, 0, 8*h0*mxx, 8*v0*myy), subsampleRatio)
    41		d.img3 = m.SubImage(image.Rect(0, 0, d.width, d.height)).(*image.YCbCr)
    42	
    43		if d.nComp == 4 {
    44			h3, v3 := d.comp[3].h, d.comp[3].v
    45			d.blackPix = make([]byte, 8*h3*mxx*8*v3*myy)
    46			d.blackStride = 8 * h3 * mxx
    47		}
    48	}
    49	
    50	// Specified in section B.2.3.
    51	func (d *decoder) processSOS(n int) error {
    52		if d.nComp == 0 {
    53			return FormatError("missing SOF marker")
    54		}
    55		if n < 6 || 4+2*d.nComp < n || n%2 != 0 {
    56			return FormatError("SOS has wrong length")
    57		}
    58		if err := d.readFull(d.tmp[:n]); err != nil {
    59			return err
    60		}
    61		nComp := int(d.tmp[0])
    62		if n != 4+2*nComp {
    63			return FormatError("SOS length inconsistent with number of components")
    64		}
    65		var scan [maxComponents]struct {
    66			compIndex uint8
    67			td        uint8 // DC table selector.
    68			ta        uint8 // AC table selector.
    69		}
    70		totalHV := 0
    71		for i := 0; i < nComp; i++ {
    72			cs := d.tmp[1+2*i] // Component selector.
    73			compIndex := -1
    74			for j, comp := range d.comp[:d.nComp] {
    75				if cs == comp.c {
    76					compIndex = j
    77				}
    78			}
    79			if compIndex < 0 {
    80				return FormatError("unknown component selector")
    81			}
    82			scan[i].compIndex = uint8(compIndex)
    83			// Section B.2.3 states that "the value of Cs_j shall be different from
    84			// the values of Cs_1 through Cs_(j-1)". Since we have previously
    85			// verified that a frame's component identifiers (C_i values in section
    86			// B.2.2) are unique, it suffices to check that the implicit indexes
    87			// into d.comp are unique.
    88			for j := 0; j < i; j++ {
    89				if scan[i].compIndex == scan[j].compIndex {
    90					return FormatError("repeated component selector")
    91				}
    92			}
    93			totalHV += d.comp[compIndex].h * d.comp[compIndex].v
    94	
    95			scan[i].td = d.tmp[2+2*i] >> 4
    96			if scan[i].td > maxTh {
    97				return FormatError("bad Td value")
    98			}
    99			scan[i].ta = d.tmp[2+2*i] & 0x0f
   100			if scan[i].ta > maxTh {
   101				return FormatError("bad Ta value")
   102			}
   103		}
   104		// Section B.2.3 states that if there is more than one component then the
   105		// total H*V values in a scan must be <= 10.
   106		if d.nComp > 1 && totalHV > 10 {
   107			return FormatError("total sampling factors too large")
   108		}
   109	
   110		// zigStart and zigEnd are the spectral selection bounds.
   111		// ah and al are the successive approximation high and low values.
   112		// The spec calls these values Ss, Se, Ah and Al.
   113		//
   114		// For progressive JPEGs, these are the two more-or-less independent
   115		// aspects of progression. Spectral selection progression is when not
   116		// all of a block's 64 DCT coefficients are transmitted in one pass.
   117		// For example, three passes could transmit coefficient 0 (the DC
   118		// component), coefficients 1-5, and coefficients 6-63, in zig-zag
   119		// order. Successive approximation is when not all of the bits of a
   120		// band of coefficients are transmitted in one pass. For example,
   121		// three passes could transmit the 6 most significant bits, followed
   122		// by the second-least significant bit, followed by the least
   123		// significant bit.
   124		//
   125		// For baseline JPEGs, these parameters are hard-coded to 0/63/0/0.
   126		zigStart, zigEnd, ah, al := int32(0), int32(blockSize-1), uint32(0), uint32(0)
   127		if d.progressive {
   128			zigStart = int32(d.tmp[1+2*nComp])
   129			zigEnd = int32(d.tmp[2+2*nComp])
   130			ah = uint32(d.tmp[3+2*nComp] >> 4)
   131			al = uint32(d.tmp[3+2*nComp] & 0x0f)
   132			if (zigStart == 0 && zigEnd != 0) || zigStart > zigEnd || blockSize <= zigEnd {
   133				return FormatError("bad spectral selection bounds")
   134			}
   135			if zigStart != 0 && nComp != 1 {
   136				return FormatError("progressive AC coefficients for more than one component")
   137			}
   138			if ah != 0 && ah != al+1 {
   139				return FormatError("bad successive approximation values")
   140			}
   141		}
   142	
   143		// mxx and myy are the number of MCUs (Minimum Coded Units) in the image.
   144		h0, v0 := d.comp[0].h, d.comp[0].v // The h and v values from the Y components.
   145		mxx := (d.width + 8*h0 - 1) / (8 * h0)
   146		myy := (d.height + 8*v0 - 1) / (8 * v0)
   147		if d.img1 == nil && d.img3 == nil {
   148			d.makeImg(mxx, myy)
   149		}
   150		if d.progressive {
   151			for i := 0; i < nComp; i++ {
   152				compIndex := scan[i].compIndex
   153				if d.progCoeffs[compIndex] == nil {
   154					d.progCoeffs[compIndex] = make([]block, mxx*myy*d.comp[compIndex].h*d.comp[compIndex].v)
   155				}
   156			}
   157		}
   158	
   159		d.bits = bits{}
   160		mcu, expectedRST := 0, uint8(rst0Marker)
   161		var (
   162			// b is the decoded coefficients, in natural (not zig-zag) order.
   163			b  block
   164			dc [maxComponents]int32
   165			// bx and by are the location of the current block, in units of 8x8
   166			// blocks: the third block in the first row has (bx, by) = (2, 0).
   167			bx, by     int
   168			blockCount int
   169		)
   170		for my := 0; my < myy; my++ {
   171			for mx := 0; mx < mxx; mx++ {
   172				for i := 0; i < nComp; i++ {
   173					compIndex := scan[i].compIndex
   174					hi := d.comp[compIndex].h
   175					vi := d.comp[compIndex].v
   176					qt := &d.quant[d.comp[compIndex].tq]
   177					for j := 0; j < hi*vi; j++ {
   178						// The blocks are traversed one MCU at a time. For 4:2:0 chroma
   179						// subsampling, there are four Y 8x8 blocks in every 16x16 MCU.
   180						//
   181						// For a baseline 32x16 pixel image, the Y blocks visiting order is:
   182						//	0 1 4 5
   183						//	2 3 6 7
   184						//
   185						// For progressive images, the interleaved scans (those with nComp > 1)
   186						// are traversed as above, but non-interleaved scans are traversed left
   187						// to right, top to bottom:
   188						//	0 1 2 3
   189						//	4 5 6 7
   190						// Only DC scans (zigStart == 0) can be interleaved. AC scans must have
   191						// only one component.
   192						//
   193						// To further complicate matters, for non-interleaved scans, there is no
   194						// data for any blocks that are inside the image at the MCU level but
   195						// outside the image at the pixel level. For example, a 24x16 pixel 4:2:0
   196						// progressive image consists of two 16x16 MCUs. The interleaved scans
   197						// will process 8 Y blocks:
   198						//	0 1 4 5
   199						//	2 3 6 7
   200						// The non-interleaved scans will process only 6 Y blocks:
   201						//	0 1 2
   202						//	3 4 5
   203						if nComp != 1 {
   204							bx = hi*mx + j%hi
   205							by = vi*my + j/hi
   206						} else {
   207							q := mxx * hi
   208							bx = blockCount % q
   209							by = blockCount / q
   210							blockCount++
   211							if bx*8 >= d.width || by*8 >= d.height {
   212								continue
   213							}
   214						}
   215	
   216						// Load the previous partially decoded coefficients, if applicable.
   217						if d.progressive {
   218							b = d.progCoeffs[compIndex][by*mxx*hi+bx]
   219						} else {
   220							b = block{}
   221						}
   222	
   223						if ah != 0 {
   224							if err := d.refine(&b, &d.huff[acTable][scan[i].ta], zigStart, zigEnd, 1<<al); err != nil {
   225								return err
   226							}
   227						} else {
   228							zig := zigStart
   229							if zig == 0 {
   230								zig++
   231								// Decode the DC coefficient, as specified in section F.2.2.1.
   232								value, err := d.decodeHuffman(&d.huff[dcTable][scan[i].td])
   233								if err != nil {
   234									return err
   235								}
   236								if value > 16 {
   237									return UnsupportedError("excessive DC component")
   238								}
   239								dcDelta, err := d.receiveExtend(value)
   240								if err != nil {
   241									return err
   242								}
   243								dc[compIndex] += dcDelta
   244								b[0] = dc[compIndex] << al
   245							}
   246	
   247							if zig <= zigEnd && d.eobRun > 0 {
   248								d.eobRun--
   249							} else {
   250								// Decode the AC coefficients, as specified in section F.2.2.2.
   251								huff := &d.huff[acTable][scan[i].ta]
   252								for ; zig <= zigEnd; zig++ {
   253									value, err := d.decodeHuffman(huff)
   254									if err != nil {
   255										return err
   256									}
   257									val0 := value >> 4
   258									val1 := value & 0x0f
   259									if val1 != 0 {
   260										zig += int32(val0)
   261										if zig > zigEnd {
   262											break
   263										}
   264										ac, err := d.receiveExtend(val1)
   265										if err != nil {
   266											return err
   267										}
   268										b[unzig[zig]] = ac << al
   269									} else {
   270										if val0 != 0x0f {
   271											d.eobRun = uint16(1 << val0)
   272											if val0 != 0 {
   273												bits, err := d.decodeBits(int32(val0))
   274												if err != nil {
   275													return err
   276												}
   277												d.eobRun |= uint16(bits)
   278											}
   279											d.eobRun--
   280											break
   281										}
   282										zig += 0x0f
   283									}
   284								}
   285							}
   286						}
   287	
   288						if d.progressive {
   289							if zigEnd != blockSize-1 || al != 0 {
   290								// We haven't completely decoded this 8x8 block. Save the coefficients.
   291								d.progCoeffs[compIndex][by*mxx*hi+bx] = b
   292								// At this point, we could execute the rest of the loop body to dequantize and
   293								// perform the inverse DCT, to save early stages of a progressive image to the
   294								// *image.YCbCr buffers (the whole point of progressive encoding), but in Go,
   295								// the jpeg.Decode function does not return until the entire image is decoded,
   296								// so we "continue" here to avoid wasted computation.
   297								continue
   298							}
   299						}
   300	
   301						// Dequantize, perform the inverse DCT and store the block to the image.
   302						for zig := 0; zig < blockSize; zig++ {
   303							b[unzig[zig]] *= qt[zig]
   304						}
   305						idct(&b)
   306						dst, stride := []byte(nil), 0
   307						if d.nComp == 1 {
   308							dst, stride = d.img1.Pix[8*(by*d.img1.Stride+bx):], d.img1.Stride
   309						} else {
   310							switch compIndex {
   311							case 0:
   312								dst, stride = d.img3.Y[8*(by*d.img3.YStride+bx):], d.img3.YStride
   313							case 1:
   314								dst, stride = d.img3.Cb[8*(by*d.img3.CStride+bx):], d.img3.CStride
   315							case 2:
   316								dst, stride = d.img3.Cr[8*(by*d.img3.CStride+bx):], d.img3.CStride
   317							case 3:
   318								dst, stride = d.blackPix[8*(by*d.blackStride+bx):], d.blackStride
   319							default:
   320								return UnsupportedError("too many components")
   321							}
   322						}
   323						// Level shift by +128, clip to [0, 255], and write to dst.
   324						for y := 0; y < 8; y++ {
   325							y8 := y * 8
   326							yStride := y * stride
   327							for x := 0; x < 8; x++ {
   328								c := b[y8+x]
   329								if c < -128 {
   330									c = 0
   331								} else if c > 127 {
   332									c = 255
   333								} else {
   334									c += 128
   335								}
   336								dst[yStride+x] = uint8(c)
   337							}
   338						}
   339					} // for j
   340				} // for i
   341				mcu++
   342				if d.ri > 0 && mcu%d.ri == 0 && mcu < mxx*myy {
   343					// A more sophisticated decoder could use RST[0-7] markers to resynchronize from corrupt input,
   344					// but this one assumes well-formed input, and hence the restart marker follows immediately.
   345					if err := d.readFull(d.tmp[:2]); err != nil {
   346						return err
   347					}
   348					if d.tmp[0] != 0xff || d.tmp[1] != expectedRST {
   349						return FormatError("bad RST marker")
   350					}
   351					expectedRST++
   352					if expectedRST == rst7Marker+1 {
   353						expectedRST = rst0Marker
   354					}
   355					// Reset the Huffman decoder.
   356					d.bits = bits{}
   357					// Reset the DC components, as per section F.2.1.3.1.
   358					dc = [maxComponents]int32{}
   359					// Reset the progressive decoder state, as per section G.1.2.2.
   360					d.eobRun = 0
   361				}
   362			} // for mx
   363		} // for my
   364	
   365		return nil
   366	}
   367	
   368	// refine decodes a successive approximation refinement block, as specified in
   369	// section G.1.2.
   370	func (d *decoder) refine(b *block, h *huffman, zigStart, zigEnd, delta int32) error {
   371		// Refining a DC component is trivial.
   372		if zigStart == 0 {
   373			if zigEnd != 0 {
   374				panic("unreachable")
   375			}
   376			bit, err := d.decodeBit()
   377			if err != nil {
   378				return err
   379			}
   380			if bit {
   381				b[0] |= delta
   382			}
   383			return nil
   384		}
   385	
   386		// Refining AC components is more complicated; see sections G.1.2.2 and G.1.2.3.
   387		zig := zigStart
   388		if d.eobRun == 0 {
   389		loop:
   390			for ; zig <= zigEnd; zig++ {
   391				z := int32(0)
   392				value, err := d.decodeHuffman(h)
   393				if err != nil {
   394					return err
   395				}
   396				val0 := value >> 4
   397				val1 := value & 0x0f
   398	
   399				switch val1 {
   400				case 0:
   401					if val0 != 0x0f {
   402						d.eobRun = uint16(1 << val0)
   403						if val0 != 0 {
   404							bits, err := d.decodeBits(int32(val0))
   405							if err != nil {
   406								return err
   407							}
   408							d.eobRun |= uint16(bits)
   409						}
   410						break loop
   411					}
   412				case 1:
   413					z = delta
   414					bit, err := d.decodeBit()
   415					if err != nil {
   416						return err
   417					}
   418					if !bit {
   419						z = -z
   420					}
   421				default:
   422					return FormatError("unexpected Huffman code")
   423				}
   424	
   425				zig, err = d.refineNonZeroes(b, zig, zigEnd, int32(val0), delta)
   426				if err != nil {
   427					return err
   428				}
   429				if zig > zigEnd {
   430					return FormatError("too many coefficients")
   431				}
   432				if z != 0 {
   433					b[unzig[zig]] = z
   434				}
   435			}
   436		}
   437		if d.eobRun > 0 {
   438			d.eobRun--
   439			if _, err := d.refineNonZeroes(b, zig, zigEnd, -1, delta); err != nil {
   440				return err
   441			}
   442		}
   443		return nil
   444	}
   445	
   446	// refineNonZeroes refines non-zero entries of b in zig-zag order. If nz >= 0,
   447	// the first nz zero entries are skipped over.
   448	func (d *decoder) refineNonZeroes(b *block, zig, zigEnd, nz, delta int32) (int32, error) {
   449		for ; zig <= zigEnd; zig++ {
   450			u := unzig[zig]
   451			if b[u] == 0 {
   452				if nz == 0 {
   453					break
   454				}
   455				nz--
   456				continue
   457			}
   458			bit, err := d.decodeBit()
   459			if err != nil {
   460				return 0, err
   461			}
   462			if !bit {
   463				continue
   464			}
   465			if b[u] >= 0 {
   466				b[u] += delta
   467			} else {
   468				b[u] -= delta
   469			}
   470		}
   471		return zig, nil
   472	}
   473	

View as plain text