// Copyright 2025 The Wuffs Authors. // // Licensed under the Apache License, Version 2.0 or the MIT license // , at your // option. This file may not be copied, modified, or distributed // except according to those terms. // // SPDX-License-Identifier: Apache-2.0 OR MIT // ---------------- // Package handsum implements the Handsum image file format. // // This is a very lossy format for very small thumbnails. Very small in terms // of image dimensions, up to 16×16 pixels, but also in terms of file size. // // The file format has three color settings (1=Gray, 3=RGB and 4=RGBA) and four // quality settings. For any given color-and-quality combination, every Handsum // image file with those settings is a fixed number of bytes. // // For color=1, also known as color=Gray: // // - A quality=1 file is 33 bytes long. // - A quality=2 file is 51 bytes long. // - A quality=3 file is 83 bytes long. // - A quality=4 file is 99 bytes long. // // For color=3, also known as color=RGB: // // - A quality=1 file is 48 bytes long. // - A quality=2 file is 75 bytes long. // - A quality=3 file is 123 bytes long. // - A quality=4 file is 147 bytes long. // // For color=4, also known as color=RGBA: // // - A quality=1 file is 72 bytes long. // - A quality=2 file is 99 bytes long. // - A quality=3 file is 147 bytes long. // - A quality=4 file is 171 bytes long. // // Every Gray/q1 image file is exactly 48 bytes (384 bits) long. For a 16×16 // pixel image, this uses 1.5 bits (0.1875 bytes) per pixel. // // Every RGB/q4 image file is exactly 147 bytes (1176 bits) long. This uses // 4.59375 bits per pixel for a 16×16 pixel image (a 1:1 aspect ratio), or // 6.125 bits per pixel for a 16×12 pixel image (a 4:3 aspect ratio). // // Handsum files start with a 3 byte header: a 15-bit magic signature, a 2-bit // color (Gray, RGB or RGBA), a 2-bit quality and a 5-bit aspect ratio. An // image's longest dimension (width or height) is 16 pixels and the aspect // ratio gives the shorter dimension. // // The color=RGB payload, after the header, holds a scaled 16×16 pixel YCbCr // 4:2:0 JPEG MCU (Minimum Coded Unit), 4 Luma and 2 Chroma blocks. Each block // is 8×8 pixels. // // The color=RGB payload ends with 2 Chroma blocks in 15, 24, 40 or 48 bytes // (depending on the quality setting). For a grayscale image, the encoded // Chroma blocks' bytes are all 0x88, and the color=Gray payload is just the 4 // Luma blocks without explicitly recording the 2 Chroma blocks. // // The color=RGBA payload extends the color=RGB payload with one more 8×8 Alpha // block. That seventh block is always encoded in 24 bytes (as if quality=4). // // Each 8×8 block is sub-divided into 8×8 (for q1), 4×4 (for q2 and q3) or 2×2 // (for q4) tiles. DCT (Discrete Cosine Transform) is applied to each tile, // producing 64, 16, 16 or 4 DCT coefficients (depending on the quality). Only // the 15 (out of 64), 6 (out of 16), 10 (out of 16) or 3 (out of 4) lowest // frequency DCT coefficients are kept for each tile. Lowest frequency means // the top-left corner in the usual visualization of JPEG's zig-zag ordering. // // In percentage terms, the quality setting keeps 23%, 38%, 63% or 75% of the // 64 DCT coefficients in each 8×8 block. // // Each DCT coefficient is encoded as one nibble (4 bits) with fixed bias and // quantization factors. // // All Handsum images use the sRGB color profile and non-premultiplied alpha. // // The "Handsum" name was inspired by the "Thumbhash" image file format (which // was in turn inspired by "Blurhash"), which is also designed for very small // thumbnails (or very compact representations of image placeholders). Handsum // files are bigger (but better quality) than Thumbhash (and Blurhash). // "Handsum" also sounds like "handsome", meaning "good looking". // // Other techniques and image formats, similar to Thumbhash, can be found by // search for "LQIP" or "Low Quality Image Placeholders". package handsum // When decoding, each nibble produces a DCT coefficient according to a simple // linear formula, with different parameters (the linear slope and intercept) // depending on DC vs AC coefficients and on the quality setting (whether the // tiles are 8×8, 4×4 or 2×2): // // ====== ======== DC ======== ======== AC ======== // Bucket q1 q2,q3 q4 q1 q2,q3 q4 // // 0x0 -1024 -512 -256 -512 -256 -128 // 0x1 -888 -444 -222 -448 -224 -112 // 0x2 -752 -376 -188 -384 -192 -96 // 0x3 -616 -308 -154 -320 -160 -80 // ... ... ... ... ... ... ... // 0x7 -72 -36 -18 -64 -32 -16 // 0x8 +64 +32 +16 0 0 0 // 0x9 +200 +100 +50 +64 +32 +16 // ... ... ... ... ... ... ... // 0xE +880 +440 +220 +384 +192 +96 // 0xF +1016 +508 +254 +448 +224 +112 // // Width 0x88 0x44 0x22 0x40 0x20 0x10 // TileSize 8 4 2 8 4 2 // ====== ======== DC ======== ======== AC ======== // // DC bucket widths are multiples of 0x11, not 0x10 with bucket=0x8 mapping to // zero, so that an input tile that is pure black (luma=0x00) or pure white // (luma=0xFF) can round-trip losslessly (as bucket 0x0 or bucket 0xF). AC maps // bucket=0x8 to zero, so that a uniformly-valued tile (all-0x00, all-0x11, // all-0x22, ..., all-0xFF) can again round-trip losslessly. // // This encoder uses smaller bucket widths for the AC coefficients, compared to // those used when decoding (42, 18, 8 instead of 64, 32, 16). This adjustment // is arbitrary but the results seem a little more vibrant. It also compensates // somewhat for the mean-reversion (gray-reversion) from the decoder's seam // smoothing, a localized blur that mitigates artifacts at tile boundaries. // // Luma values can span the full range [0x00, 0xFF] but decoded Chroma values // are restricted to a narrower range [0x3C, 0xBB], as extreme Chroma values // are rare in practice. The latter range has a slight bias (not centred on the // neutral 0x80) so that the Chroma DC components of a neutral gray image can // round-trip losslessly (in the 0x8 bucket). Chroma is narrowed (after Inverse // DCT) on decode and so is widened (before Forward DCT) on encode. Luma, // wide-Chroma and Alpha then use the same DCT buckets. // // For Alpha values only, to encourage producing 0x00 (fully transparent) or // 0xFF (fully opaque) values, we transform (on decode) the values towards the // extreme ends via a cube root mapping. // // There are 16 possible landscape aspect ratios (and likewise for portrait): // // 16 : 1 ≈ 16.000 : 1 // 16 : 2 ≈ 8.000 : 1 // 16 : 3 ≈ 5.333 : 1 // 16 : 4 ≈ 4.000 : 1 // 16 : 5 ≈ 3.200 : 1 // 16 : 6 ≈ 2.667 : 1 // 16 : 7 ≈ 2.286 : 1 // 16 : 8 ≈ 2.000 : 1 // 16 : 9 ≈ 1.778 : 1 // 16 : 10 ≈ 1.600 : 1 // 16 : 11 ≈ 1.455 : 1 // 16 : 12 ≈ 1.333 : 1 // 16 : 13 ≈ 1.231 : 1 // 16 : 14 ≈ 1.143 : 1 // 16 : 15 ≈ 1.067 : 1 // 16 : 16 ≈ 1.000 : 1 import ( "errors" "image" "image/color" "io" "github.com/google/wuffs/lib/lowleveldct4x4" "github.com/google/wuffs/lib/lowleveljpeg" "golang.org/x/image/draw" ) // MaxDimension is the maximum (inclusive) width or height of every Handsum // image file. // // Every image is either (W × 16) or (16 × H) or both, for some positive W or H // that is no greater than 16. const MaxDimension = 16 // MagicEtc is the byte string prefix of every Handsum image file. A Gray image // starts with Magic0. An RGB or RGBA image starts with Magic1. The two 16-bit // strings only differ in their final bit. // // It's like how every JPEG image file starts with "\xFF\xD8". const ( Magic0 = "\xFE\xD6" Magic1 = "\xFE\xD7" ) func init() { image.RegisterFormat("handsum", Magic0, Decode, DecodeConfig) image.RegisterFormat("handsum", Magic1, Decode, DecodeConfig) } var ( ErrBadArgument = errors.New("handsum: bad argument") ErrNotAHandsumFile = errors.New("handsum: not a handsum file") ) const ( fileSizeHeader = 3 fileSizeMax = 171 ) func fileSize(c Color, q Quality) int { // 33, 51, 83, 99. // 0, 0, 0, 0. // 48, 75, 123, 147. // 72, 99, 147, 171. return int(("" + "\x21\x33\x53\x63" + "\x00\x00\x00\x00" + "\x30\x4B\x7B\x93" + "\x48\x63\x93\xAB")[(int((c-1)&3)<<2)|int((q-1)&3)]) } // Color is a Handsum image's color setting, either 1 (Gray), 3 (RGB) or 4 // (RGBA). type Color uint8 const ( ColorGray = Color(1) ColorRGB = Color(3) ColorRGBA = Color(4) ) func (c Color) numberOfBlocks() int { if c == ColorGray { return 4 } return 6 } // Quality is a Handsum image's quality setting, from 1 (worst) to 4 (best). // // "Best" is relative to the other settings. In absolute terms, Handsum's image // quality ranges from "potato" (best) to "extremely potato" (worst). type Quality uint8 const ( QualityWorst = Quality(1) QualityMediumLow = Quality(2) QualityMediumHigh = Quality(3) QualityBest = Quality(4) ) func (q Quality) numberOfCoefficients() int { // 15 (out of 64), 6 (out of 16), 10 (out of 16), 3 (out of 4). return int("\x0F\x06\x0A\x03"[(q-1)&3]) } // EncodeOptions are optional arguments to Encode. The zero value is valid and // means to use the default configuration, encoding to 147 bytes. type EncodeOptions struct { // Color is the color setting. The zero value means to use the default, // ColorRGB. Color Color // Quality is the quality-versus-file-size setting. The zero value means to // use the default, QualityBest (which is also the largest file size). Quality Quality } func (o *EncodeOptions) color() Color { if o != nil { switch o.Color { case ColorGray, ColorRGB, ColorRGBA: return o.Color } } return ColorRGB } func (o *EncodeOptions) quality() Quality { if o != nil { switch o.Quality { case QualityWorst, QualityMediumLow, QualityMediumHigh, QualityBest: return o.Quality } } return QualityBest } // Encode writes src to w in the Handsum format. // // options may be nil, which means to use the default configuration. func Encode(w io.Writer, src image.Image, options *EncodeOptions) error { if (w == nil) || (src == nil) { return ErrBadArgument } srcB := src.Bounds() srcW, srcH := srcB.Dx(), srcB.Dy() if (srcW <= 0) || (srcH <= 0) { return ErrBadArgument } aspectRatio := byte(0) if srcW >= srcH { // Landscape. a := ((int64(srcH) * 32) + int64(srcW)) / (2 * int64(srcW)) if a <= 0 { a = 1 } aspectRatio = byte(a-1) | 0x00 } else { // Portrait. a := ((int64(srcW) * 32) + int64(srcH)) / (2 * int64(srcH)) if a <= 0 { a = 1 } aspectRatio = byte(a-1) | 0x10 if aspectRatio == 0x1F { // Reserved for future expansion. aspectRatio = 0x0F } } alphasQuadBlock := lowleveljpeg.QuadBlockU8{} dst := scaleSrc(src, &alphasQuadBlock) dstU8s := lowleveljpeg.Array6BlockU8{} dstU8s.ExtractYCbCrFrom(dst, 0, 0) // Biasing the Chroma blocks by +8 shifts the neutral (gray) Chroma values // from 0x80 to 0x88. DC coefficient bias-and-quantization can losslessly // encode a block whose elements are a uniform multiple of 0x11. scaleAndBiasChromaUp(&dstU8s[4]) scaleAndBiasChromaUp(&dstU8s[5]) c := options.color() q := options.quality() buf := [fileSizeMax]byte{} buf[0] = Magic0[0] buf[1] = Magic0[1] | uint8((c-1)>>1) buf[2] = ((uint8(c - 1)) << 7) | ((uint8(q - 1)) << 5) | aspectRatio bitOffset := 3 * 8 encodeBlock := encodeBlockFuncs[(q-1)&3] for i := range c.numberOfBlocks() { bitOffset = encodeBlock(&buf, bitOffset, &dstU8s[i], q.numberOfCoefficients()) } if c >= ColorRGBA { alphasBlock := lowleveljpeg.BlockU8{} alphasBlock.DownsampleFrom(&alphasQuadBlock) bitOffset = encodeBlockQ4(&buf, bitOffset, &alphasBlock, QualityBest.numberOfCoefficients()) } _, err := w.Write(buf[:bitOffset/8]) return err } func scaleSrc(src image.Image, alphasQuadBlock *lowleveljpeg.QuadBlockU8) image.Image { if o, ok := src.(interface{ Opaque() bool }); ok && o.Opaque() { for i := range alphasQuadBlock { alphasQuadBlock[i] = 0xFF } dst := image.NewRGBA(image.Rectangle{Max: image.Point{X: 16, Y: 16}}) draw.BiLinear.Scale(dst, dst.Bounds(), src, src.Bounds(), draw.Src, nil) return dst } dst := image.NewNRGBA(image.Rectangle{Max: image.Point{X: 16, Y: 16}}) draw.BiLinear.Scale(dst, dst.Bounds(), src, src.Bounds(), draw.Src, nil) // Extract alphasQuadBlock (as a side effect of calling this function) and // set dst's alpha values to 0xFF. for i := range alphasQuadBlock { v := dst.Pix[(4*i)+3] dst.Pix[(4*i)+3] = 0xFF alphasQuadBlock[i] = v } return &image.RGBA{ // Re-interpret NRGBA as RGBA. Pix: dst.Pix, Stride: dst.Stride, Rect: dst.Rect, } } type encodeBlockFunc func(buf *[fileSizeMax]byte, bitOffset int, src *lowleveljpeg.BlockU8, nCoeffs int) int var encodeBlockFuncs = [4]encodeBlockFunc{ encodeBlockQ1, encodeBlockQ2, encodeBlockQ2, encodeBlockQ4, } func encodeBlockQ1(buf *[fileSizeMax]byte, bitOffset int, src *lowleveljpeg.BlockU8, nCoeffs int) int { f8 := src.ForwardDCT() for _, z := range zigzag8 { v := int32(f8[z]) if z == 0 { v = (v + 1024 + 0x44) / 0x88 } else { v = (v + (8.5 * 42)) / 42 } buf[bitOffset>>3] |= nibblify(v) << (bitOffset & 4) bitOffset += 4 } return bitOffset } func encodeBlockQ2(buf *[fileSizeMax]byte, bitOffset int, src *lowleveljpeg.BlockU8, nCoeffs int) int { for i := range 4 { x4 := (i & 1) << 2 // The sequence 0, 4, 0, 4. y4 := (i & 2) << 1 // The sequence 0, 0, 4, 4. o4 := (8 * y4) + x4 // The sequence 0, 4, 32, 36. b4 := extract4x4Tile(src, o4) f4 := b4.ForwardDCT() for _, z := range zigzag4[:nCoeffs] { v := int32(f4[z]) if z == 0 { v = (v + 512 + 0x22) / 0x44 } else { v = (v + (8.5 * 18)) / 18 } buf[bitOffset>>3] |= nibblify(v) << (bitOffset & 4) bitOffset += 4 } } return bitOffset } func encodeBlockQ4(buf *[fileSizeMax]byte, bitOffset int, src *lowleveljpeg.BlockU8, nCoeffs int) int { // We could import the lowleveldct2x2 package and call its ForwardDCT // method (and InverseDCTFrom on decode), but it's easy to just inline it. for i := range 16 { x := 2 * (i & 3) y := 2 * (i >> 2) j := (8 * y) + x s0 := int32(src[j+0]) - 0x80 s1 := int32(src[j+1]) - 0x80 s2 := int32(src[j+8]) - 0x80 s3 := int32(src[j+9]) - 0x80 dct0 := (+s0 + s1 + s2 + s3 + 1) >> 1 dct1 := (+s0 - s1 + s2 - s3 + 1) >> 1 dct2 := (+s0 + s1 - s2 - s3 + 1) >> 1 v0 := (dct0 + 256 + 0x11) / 0x22 buf[bitOffset>>3] |= nibblify(v0) << (bitOffset & 4) bitOffset += 4 v1 := (dct1 + (8.5 * 8)) / 8 buf[bitOffset>>3] |= nibblify(v1) << (bitOffset & 4) bitOffset += 4 v2 := (dct2 + (8.5 * 8)) / 8 buf[bitOffset>>3] |= nibblify(v2) << (bitOffset & 4) bitOffset += 4 } return bitOffset } // DecodeConfig reads a Handsum image configuration from r. func DecodeConfig(r io.Reader) (image.Config, error) { buf := [fileSizeHeader]byte{} if _, err := io.ReadFull(r, buf[:]); err != nil { return image.Config{}, err } else if (buf[0] != Magic0[0]) || ((0xFE & buf[1]) != Magic0[1]) { return image.Config{}, ErrNotAHandsumFile } c, ok := decodeColorSetting(buf[1], buf[2]) if !ok { return image.Config{}, ErrNotAHandsumFile } w, h, ok := decodeWidthAndHeight(buf[2]) if !ok { return image.Config{}, ErrNotAHandsumFile } cm := color.GrayModel if c == ColorRGB { cm = color.RGBAModel } else if c == ColorRGBA { cm = color.NRGBAModel } return image.Config{ ColorModel: cm, Width: w, Height: h, }, nil } func decodeColorSetting(buf1 byte, buf2 byte) (Color, bool) { switch ((buf1 & 1) << 1) | (buf2 >> 7) { case 0: return ColorGray, true case 2: return ColorRGB, true case 3: return ColorRGBA, true } return 0, false } // Decode reads a Handsum image from r. func Decode(r io.Reader) (image.Image, error) { buf := [fileSizeMax]byte{} if _, err := io.ReadFull(r, buf[:fileSizeHeader]); err != nil { return nil, err } else if (buf[0] != Magic0[0]) || ((0xFE & buf[1]) != Magic0[1]) { return nil, ErrNotAHandsumFile } c, ok := decodeColorSetting(buf[1], buf[2]) if !ok { return nil, ErrNotAHandsumFile } q := Quality((buf[2]>>5)&0x03) + 1 if _, err := io.ReadFull(r, buf[fileSizeHeader:fileSize(c, q)]); err != nil { return nil, err } dstW, dstH, ok := decodeWidthAndHeight(buf[2]) if !ok { return nil, ErrNotAHandsumFile } bitOffset := 3 * 8 decodeBlock := decodeBlockFuncs[(q-1)&3] lumaQuadBlockU8 := lowleveljpeg.QuadBlockU8{} bitOffset = decodeBlock(lumaQuadBlockU8[0x00:], 16, &buf, bitOffset, q.numberOfCoefficients()) bitOffset = decodeBlock(lumaQuadBlockU8[0x08:], 16, &buf, bitOffset, q.numberOfCoefficients()) bitOffset = decodeBlock(lumaQuadBlockU8[0x80:], 16, &buf, bitOffset, q.numberOfCoefficients()) bitOffset = decodeBlock(lumaQuadBlockU8[0x88:], 16, &buf, bitOffset, q.numberOfCoefficients()) smoothBlockSeams16x16(&lumaQuadBlockU8) cbQuadBlockU8 := lowleveljpeg.QuadBlockU8{} crQuadBlockU8 := lowleveljpeg.QuadBlockU8{} aaQuadBlockU8 := lowleveljpeg.QuadBlockU8{} for i := range aaQuadBlockU8 { aaQuadBlockU8[i] = 0xFF } if c >= ColorRGB { cbBlockU8 := lowleveljpeg.BlockU8{} crBlockU8 := lowleveljpeg.BlockU8{} bitOffset = decodeBlock(cbBlockU8[:], 8, &buf, bitOffset, q.numberOfCoefficients()) bitOffset = decodeBlock(crBlockU8[:], 8, &buf, bitOffset, q.numberOfCoefficients()) scaleAndBiasChromaDown(&cbBlockU8) scaleAndBiasChromaDown(&crBlockU8) cbQuadBlockU8.UpsampleFrom(&cbBlockU8) crQuadBlockU8.UpsampleFrom(&crBlockU8) if c >= ColorRGBA { aaBlockU8 := lowleveljpeg.BlockU8{} bitOffset = decodeBlockQ4(aaBlockU8[:], 8, &buf, bitOffset, QualityBest.numberOfCoefficients()) aaQuadBlockU8.UpsampleFrom(&aaBlockU8) for i, v := range aaQuadBlockU8 { aaQuadBlockU8[i] = cubeRootLUT[v] } } } return finishDecode(dstW, dstH, c, &lumaQuadBlockU8, &cbQuadBlockU8, &crQuadBlockU8, &aaQuadBlockU8), nil } func decodeWidthAndHeight(buf2 byte) (w int, h int, ok bool) { if (buf2 & 0x1F) == 0x1F { return 0, 0, false } else if (buf2 & 0x10) == 0x00 { // Landscape. w = 16 h = 1 + int(buf2&0x0F) } else { // Portrait. w = 1 + int(buf2&0x0F) h = 16 } return w, h, true } type decodeBlockFunc func(dst []byte, stride int, buf *[fileSizeMax]byte, bitOffset int, nCoeffs int) int var decodeBlockFuncs = [4]decodeBlockFunc{ decodeBlockQ1, decodeBlockQ2, decodeBlockQ2, decodeBlockQ4, } func decodeBlockQ1(dst []byte, stride int, buf *[fileSizeMax]byte, bitOffset int, nCoeffs int) int { a := lowleveljpeg.BlockI16{} for _, z := range zigzag8 { nibble := (buf[bitOffset>>3] >> (bitOffset & 4)) & 15 if z == 0 { a[z] = int16((int32(nibble) * 0x88) - 1024) } else { a[z] = int16((int32(nibble) * 0x40) - 512) } bitOffset += 4 } b := lowleveljpeg.BlockU8{} b.InverseDCTFrom(&a) for i := range 8 { di := i * stride bi := i * 8 copy(dst[di:di+8], b[bi:bi+8]) } return bitOffset } func decodeBlockQ2(dst []byte, stride int, buf *[fileSizeMax]byte, bitOffset int, nCoeffs int) int { tmp := lowleveljpeg.BlockU8{} for i := range 4 { a := lowleveldct4x4.BlockI16{} for _, z := range zigzag4[:nCoeffs] { nibble := (buf[bitOffset>>3] >> (bitOffset & 4)) & 15 if z == 0 { a[z] = int16((int32(nibble) * 0x44) - 512) } else { a[z] = int16((int32(nibble) * 0x20) - 256) } bitOffset += 4 } b := lowleveldct4x4.BlockU8{} b.InverseDCTFrom(&a) x4 := (i & 1) << 2 // The sequence 0, 4, 0, 4. y4 := (i & 2) << 1 // The sequence 0, 0, 4, 4. for i := range 4 { ti := ((i + y4) * 8) + x4 bi := i * 4 copy(tmp[ti:ti+4], b[bi:bi+4]) } } smoothBlockSeams8x8Q2(&tmp) for i := range 8 { di := i * stride ti := i * 8 copy(dst[di:di+8], tmp[ti:ti+8]) } return bitOffset } func decodeBlockQ4(dst []byte, stride int, buf *[fileSizeMax]byte, bitOffset int, nCoeffs int) int { tmp := lowleveljpeg.BlockU8{} for i := range 16 { nibble0 := (buf[bitOffset>>3] >> (bitOffset & 4)) & 15 dct0 := (int32(nibble0) * 0x22) - 256 bitOffset += 4 nibble1 := (buf[bitOffset>>3] >> (bitOffset & 4)) & 15 dct1 := (int32(nibble1) * 0x10) - 128 bitOffset += 4 nibble2 := (buf[bitOffset>>3] >> (bitOffset & 4)) & 15 dct2 := (int32(nibble2) * 0x10) - 128 bitOffset += 4 x2 := 2 * (i & 3) // The sequence 0, 2, 4, 6, 0, 2, 4, ..., 6. y2 := 2 * (i >> 2) // The sequence 0, 0, 0, 0, 2, 2, 2, ..., 6. o2 := (y2 * 8) + x2 // The sequence 0, 2, 4, 6, 16, 18, 20, ..., 54. tmp[o2+0] = biasAndClamp[((+dct0+dct1+dct2+1)>>1)&1023] tmp[o2+1] = biasAndClamp[((+dct0-dct1+dct2+1)>>1)&1023] tmp[o2+8] = biasAndClamp[((+dct0+dct1-dct2+1)>>1)&1023] tmp[o2+9] = biasAndClamp[((+dct0-dct1-dct2+1)>>1)&1023] } smoothBlockSeams8x8Q4(&tmp) for i := range 8 { di := i * stride ti := i * 8 copy(dst[di:di+8], tmp[ti:ti+8]) } return bitOffset } func finishDecode(w int, h int, c Color, yy *lowleveljpeg.QuadBlockU8, cb *lowleveljpeg.QuadBlockU8, cr *lowleveljpeg.QuadBlockU8, aa *lowleveljpeg.QuadBlockU8) image.Image { tmp := [1024]byte{} if c == ColorGray { for i := range 256 { tmp[(4 * i)] = yy[i] } } else { for i := range 256 { tmp[(4*i)+0], tmp[(4*i)+1], tmp[(4*i)+2] = color.YCbCrToRGB(yy[i], cb[i], cr[i]) tmp[(4*i)+3] = aa[i] } } pix := tmp[:] if w < 16 { pix = scaleHorizontal(&tmp, uint32(w)) } else if h < 16 { pix = scaleVertical(&tmp, uint32(h)) } if c == ColorGray { return &image.Gray{ Pix: convertYxxxToY(pix), Stride: w, Rect: image.Rect(0, 0, w, h), } } else if c == ColorRGB { return &image.RGBA{ Pix: pix, Stride: 4 * w, Rect: image.Rect(0, 0, w, h), } } return &image.NRGBA{ Pix: pix, Stride: 4 * w, Rect: image.Rect(0, 0, w, h), } } func scaleHorizontal(src *[1024]byte, w uint32) []byte { dst := make([]byte, 64*w) for y := range 16 { dstx := 0 acc0 := uint32(0) acc1 := uint32(0) acc2 := uint32(0) acc3 := uint32(0) remainder := uint32(16) for srcx := range 16 { si := (64 * y) + (4 * srcx) s0 := uint32(src[si+0]) s1 := uint32(src[si+1]) s2 := uint32(src[si+2]) s3 := uint32(src[si+3]) if remainder > w { remainder -= w acc0 += w * s0 acc1 += w * s1 acc2 += w * s2 acc3 += w * s3 } else { acc0 += remainder * s0 acc1 += remainder * s1 acc2 += remainder * s2 acc3 += remainder * s3 di := (4 * int(w) * y) + (4 * dstx) dst[di+0] = uint8((acc0 + 8) / 16) dst[di+1] = uint8((acc1 + 8) / 16) dst[di+2] = uint8((acc2 + 8) / 16) dst[di+3] = uint8((acc3 + 8) / 16) dstx++ partial := w - remainder acc0 = partial * s0 acc1 = partial * s1 acc2 = partial * s2 acc3 = partial * s3 remainder = 16 - partial } } } return dst } func scaleVertical(src *[1024]byte, h uint32) []byte { dst := make([]byte, 64*h) for x := range 16 { dsty := 0 acc0 := uint32(0) acc1 := uint32(0) acc2 := uint32(0) acc3 := uint32(0) remainder := uint32(16) for srcy := range 16 { si := (64 * srcy) + (4 * x) s0 := uint32(src[si+0]) s1 := uint32(src[si+1]) s2 := uint32(src[si+2]) s3 := uint32(src[si+3]) if remainder > h { remainder -= h acc0 += h * s0 acc1 += h * s1 acc2 += h * s2 acc3 += h * s3 } else { acc0 += remainder * s0 acc1 += remainder * s1 acc2 += remainder * s2 acc3 += remainder * s3 di := (64 * dsty) + (4 * x) dst[di+0] = uint8((acc0 + 8) / 16) dst[di+1] = uint8((acc1 + 8) / 16) dst[di+2] = uint8((acc2 + 8) / 16) dst[di+3] = uint8((acc3 + 8) / 16) dsty++ partial := h - remainder acc0 = partial * s0 acc1 = partial * s1 acc2 = partial * s2 acc3 = partial * s3 remainder = 16 - partial } } } return dst } func convertYxxxToY(pix []byte) []byte { ret := make([]byte, len(pix)/4) for i := range ret { ret[i] = pix[4*i] } return ret } // zigzag8 represents JPEG's zig-zag order for visiting DCT coefficients. // QualityWorst only uses the first (1 + 2 + 3 + 4 + 5) = 15 of JPEG's 64 DCT // coefficients. // // https://en.wikipedia.org/wiki/File:JPEG_ZigZag.svg var zigzag8 = [15]uint8{ 0o00, 0o01, 0o10, 0o20, 0o11, 0o02, 0o03, 0o12, // 0, 1, 8, 16, 9, 2, 3, 10, 0o21, 0o30, 0o40, 0o31, 0o22, 0o13, 0o04, // 17, 24, 32, 25, 18, 11, 4, } // zigzag4 is like zigzag8 but using the first (1 + 2 + 3) = 6 or (1 + 2 + 3 + // 4) = 10 (instead of 15) DCT coefficients of a 4×4 (instead of 8×8) block. var zigzag4 = [10]uint8{ 0o00, 0o01, 0o04, 0o10, 0o05, 0o02, 0o03, 0o06, // 0, 1, 4, 8, 5, 2, 3, 6, 0o11, 0o14, // 9, 12, } // nibblify clamps v to the range [0x0, 0xF]. func nibblify(v int32) uint8 { return uint8(max(0x0, min(0xF, v))) } // extract4x4Tile extracts a 4×4 tile out of an 8×8 block. func extract4x4Tile(b8 *lowleveljpeg.BlockU8, offset int) lowleveldct4x4.BlockU8 { return lowleveldct4x4.BlockU8{ b8[offset+0x00], b8[offset+0x01], b8[offset+0x02], b8[offset+0x03], b8[offset+0x08], b8[offset+0x09], b8[offset+0x0A], b8[offset+0x0B], b8[offset+0x10], b8[offset+0x11], b8[offset+0x12], b8[offset+0x13], b8[offset+0x18], b8[offset+0x19], b8[offset+0x1A], b8[offset+0x1B], } } func smoothBlockSeams16x16(b *lowleveljpeg.QuadBlockU8) { for _, pair := range smoothingPairs16x16 { v0 := uint32(b[pair[0]]) v1 := uint32(b[pair[1]]) b[pair[0]] = uint8(((3 * v0) + v1 + 2) / 4) b[pair[1]] = uint8(((3 * v1) + v0 + 2) / 4) } v77 := uint32(b[0x77]) v78 := uint32(b[0x78]) v88 := uint32(b[0x88]) v87 := uint32(b[0x87]) b[0x77] = uint8(((9 * v77) + (3 * v78) + v88 + (3 * v87) + 8) / 16) b[0x78] = uint8(((9 * v78) + (3 * v88) + v87 + (3 * v77) + 8) / 16) b[0x88] = uint8(((9 * v88) + (3 * v87) + v77 + (3 * v78) + 8) / 16) b[0x87] = uint8(((9 * v87) + (3 * v77) + v78 + (3 * v88) + 8) / 16) } func smoothBlockSeams8x8Q2(b *lowleveljpeg.BlockU8) { for _, pair := range smoothingPairs8x8 { v0 := uint32(b[pair[0]]) v1 := uint32(b[pair[1]]) b[pair[0]] = uint8(((3 * v0) + v1 + 2) / 4) b[pair[1]] = uint8(((3 * v1) + v0 + 2) / 4) } v33 := uint32(b[0o33]) v34 := uint32(b[0o34]) v44 := uint32(b[0o44]) v43 := uint32(b[0o43]) b[0o33] = uint8(((9 * v33) + (3 * v34) + v44 + (3 * v43) + 8) / 16) b[0o34] = uint8(((9 * v34) + (3 * v44) + v43 + (3 * v33) + 8) / 16) b[0o44] = uint8(((9 * v44) + (3 * v43) + v33 + (3 * v34) + 8) / 16) b[0o43] = uint8(((9 * v43) + (3 * v33) + v34 + (3 * v44) + 8) / 16) } func smoothBlockSeams8x8Q4(b *lowleveljpeg.BlockU8) { for y := 1; y < 7; y += 2 { for x := 1; x < 7; x += 2 { o := (y * 8) + x v0 := uint32(b[o+0]) v1 := uint32(b[o+1]) v9 := uint32(b[o+9]) v8 := uint32(b[o+8]) b[o+0] = uint8(((9 * v0) + (3 * v1) + v9 + (3 * v8) + 8) / 16) b[o+1] = uint8(((9 * v1) + (3 * v9) + v8 + (3 * v0) + 8) / 16) b[o+9] = uint8(((9 * v9) + (3 * v8) + v0 + (3 * v1) + 8) / 16) b[o+8] = uint8(((9 * v8) + (3 * v0) + v1 + (3 * v9) + 8) / 16) } { v0 := uint32(b[0o00+y]) v1 := uint32(b[0o01+y]) b[0o00+y] = uint8(((3 * v0) + v1 + 2) / 4) b[0o01+y] = uint8(((3 * v1) + v0 + 2) / 4) } { v0 := uint32(b[0o70+y]) v1 := uint32(b[0o71+y]) b[0o70+y] = uint8(((3 * v0) + v1 + 2) / 4) b[0o71+y] = uint8(((3 * v1) + v0 + 2) / 4) } y8 := y * 8 { v0 := uint32(b[0o00+y8]) v1 := uint32(b[0o10+y8]) b[0o00+y8] = uint8(((3 * v0) + v1 + 2) / 4) b[0o10+y8] = uint8(((3 * v1) + v0 + 2) / 4) } { v0 := uint32(b[0o07+y8]) v1 := uint32(b[0o17+y8]) b[0o07+y8] = uint8(((3 * v0) + v1 + 2) / 4) b[0o17+y8] = uint8(((3 * v1) + v0 + 2) / 4) } } } // smoothingPairs16x16 are the seams of the four 8×8 Luma blocks in a 16×16 // MCU. The central 4 pixels are handled separately. var smoothingPairs16x16 = [28][2]uint8{ {0x07, 0x08}, {0x17, 0x18}, {0x27, 0x28}, {0x37, 0x38}, {0x47, 0x48}, {0x57, 0x58}, {0x67, 0x68}, {0x70, 0x80}, {0x71, 0x81}, {0x72, 0x82}, {0x73, 0x83}, {0x74, 0x84}, {0x75, 0x85}, {0x76, 0x86}, {0x79, 0x89}, {0x7A, 0x8A}, {0x7B, 0x8B}, {0x7C, 0x8C}, {0x7D, 0x8D}, {0x7E, 0x8E}, {0x7F, 0x8F}, {0x97, 0x98}, {0xA7, 0xA8}, {0xB7, 0xB8}, {0xC7, 0xC8}, {0xD7, 0xD8}, {0xE7, 0xE8}, {0xF7, 0xF8}, } // smoothingPairs8x8 is like smoothingPairs16x16 but for the seams of the four // 4×4 quadrants of an 8×8. var smoothingPairs8x8 = [12][2]uint8{ {0o03, 0o04}, {0o13, 0o14}, {0o23, 0o24}, {0o30, 0o40}, {0o31, 0o41}, {0o32, 0o42}, {0o35, 0o45}, {0o36, 0o46}, {0o37, 0o47}, {0o53, 0o54}, {0o63, 0o64}, {0o73, 0o74}, } // scaleAndBiasChromaUp is equivalent to this [256]uint8 look-up table. // // 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x04, 0x06, // 0x08, 0x0A, 0x0C, 0x0E, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1A, 0x1C, 0x1E, 0x20, 0x22, 0x24, 0x26, // 0x28, 0x2A, 0x2C, 0x2E, 0x30, 0x32, 0x34, 0x36, 0x38, 0x3A, 0x3C, 0x3E, 0x40, 0x42, 0x44, 0x46, // 0x48, 0x4A, 0x4C, 0x4E, 0x50, 0x52, 0x54, 0x56, 0x58, 0x5A, 0x5C, 0x5E, 0x60, 0x62, 0x64, 0x66, // 0x68, 0x6A, 0x6C, 0x6E, 0x70, 0x72, 0x74, 0x76, 0x78, 0x7A, 0x7C, 0x7E, 0x80, 0x82, 0x84, 0x86, // // 0x88, 0x8A, 0x8C, 0x8E, 0x90, 0x92, 0x94, 0x96, 0x98, 0x9A, 0x9C, 0x9E, 0xA0, 0xA2, 0xA4, 0xA6, // 0xA8, 0xAA, 0xAC, 0xAE, 0xB0, 0xB2, 0xB4, 0xB6, 0xB8, 0xBA, 0xBC, 0xBE, 0xC0, 0xC2, 0xC4, 0xC6, // 0xC8, 0xCA, 0xCC, 0xCE, 0xD0, 0xD2, 0xD4, 0xD6, 0xD8, 0xDA, 0xDC, 0xDE, 0xE0, 0xE2, 0xE4, 0xE6, // 0xE8, 0xEA, 0xEC, 0xEE, 0xF0, 0xF2, 0xF4, 0xF6, 0xF8, 0xFA, 0xFC, 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, // 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, func scaleAndBiasChromaUp(b *lowleveljpeg.BlockU8) { for i, v := range b { b[i] = uint8(max(0x00, min(0xFF, ((int(v)*2)-0x78)))) } } // scaleAndBiasChromaDown is equivalent to this [256]uint8 look-up table. // // 0x3C, 0x3C, 0x3D, 0x3D, 0x3E, 0x3E, 0x3F, 0x3F, 0x40, 0x40, 0x41, 0x41, 0x42, 0x42, 0x43, 0x43, // 0x44, 0x44, 0x45, 0x45, 0x46, 0x46, 0x47, 0x47, 0x48, 0x48, 0x49, 0x49, 0x4A, 0x4A, 0x4B, 0x4B, // 0x4C, 0x4C, 0x4D, 0x4D, 0x4E, 0x4E, 0x4F, 0x4F, 0x50, 0x50, 0x51, 0x51, 0x52, 0x52, 0x53, 0x53, // 0x54, 0x54, 0x55, 0x55, 0x56, 0x56, 0x57, 0x57, 0x58, 0x58, 0x59, 0x59, 0x5A, 0x5A, 0x5B, 0x5B, // 0x5C, 0x5C, 0x5D, 0x5D, 0x5E, 0x5E, 0x5F, 0x5F, 0x60, 0x60, 0x61, 0x61, 0x62, 0x62, 0x63, 0x63, // 0x64, 0x64, 0x65, 0x65, 0x66, 0x66, 0x67, 0x67, 0x68, 0x68, 0x69, 0x69, 0x6A, 0x6A, 0x6B, 0x6B, // 0x6C, 0x6C, 0x6D, 0x6D, 0x6E, 0x6E, 0x6F, 0x6F, 0x70, 0x70, 0x71, 0x71, 0x72, 0x72, 0x73, 0x73, // 0x74, 0x74, 0x75, 0x75, 0x76, 0x76, 0x77, 0x77, 0x78, 0x78, 0x79, 0x79, 0x7A, 0x7A, 0x7B, 0x7B, // // 0x7C, 0x7C, 0x7D, 0x7D, 0x7E, 0x7E, 0x7F, 0x7F, 0x80, 0x80, 0x81, 0x81, 0x82, 0x82, 0x83, 0x83, // 0x84, 0x84, 0x85, 0x85, 0x86, 0x86, 0x87, 0x87, 0x88, 0x88, 0x89, 0x89, 0x8A, 0x8A, 0x8B, 0x8B, // 0x8C, 0x8C, 0x8D, 0x8D, 0x8E, 0x8E, 0x8F, 0x8F, 0x90, 0x90, 0x91, 0x91, 0x92, 0x92, 0x93, 0x93, // 0x94, 0x94, 0x95, 0x95, 0x96, 0x96, 0x97, 0x97, 0x98, 0x98, 0x99, 0x99, 0x9A, 0x9A, 0x9B, 0x9B, // 0x9C, 0x9C, 0x9D, 0x9D, 0x9E, 0x9E, 0x9F, 0x9F, 0xA0, 0xA0, 0xA1, 0xA1, 0xA2, 0xA2, 0xA3, 0xA3, // 0xA4, 0xA4, 0xA5, 0xA5, 0xA6, 0xA6, 0xA7, 0xA7, 0xA8, 0xA8, 0xA9, 0xA9, 0xAA, 0xAA, 0xAB, 0xAB, // 0xAC, 0xAC, 0xAD, 0xAD, 0xAE, 0xAE, 0xAF, 0xAF, 0xB0, 0xB0, 0xB1, 0xB1, 0xB2, 0xB2, 0xB3, 0xB3, // 0xB4, 0xB4, 0xB5, 0xB5, 0xB6, 0xB6, 0xB7, 0xB7, 0xB8, 0xB8, 0xB9, 0xB9, 0xBA, 0xBA, 0xBB, 0xBB, func scaleAndBiasChromaDown(b *lowleveljpeg.BlockU8) { for i, v := range b { b[i] = (v >> 1) + 0x3C } } // cubeRootLUT[x] is the cube root of x, when mapping the uint8 range [0x00, // 0xFF] to float64 range [-128/128, +127/128]. var cubeRootLUT = [256]uint8{ 0x00, 0x00, 0x01, 0x01, 0x01, 0x02, 0x02, 0x02, 0x03, 0x03, 0x03, 0x04, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06, 0x06, 0x07, 0x07, 0x07, 0x08, 0x08, 0x09, 0x09, 0x09, 0x0A, 0x0A, 0x0B, 0x0B, 0x0B, 0x0C, 0x0C, 0x0D, 0x0D, 0x0D, 0x0E, 0x0E, 0x0F, 0x0F, 0x0F, 0x10, 0x10, 0x11, 0x11, 0x12, 0x12, 0x13, 0x13, 0x13, 0x14, 0x14, 0x15, 0x15, 0x16, 0x16, 0x17, 0x17, 0x18, 0x18, 0x19, 0x19, 0x1A, 0x1A, 0x1B, 0x1B, 0x1C, 0x1D, 0x1D, 0x1E, 0x1E, 0x1F, 0x1F, 0x20, 0x21, 0x21, 0x22, 0x22, 0x23, 0x24, 0x24, 0x25, 0x26, 0x26, 0x27, 0x28, 0x28, 0x29, 0x2A, 0x2B, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F, 0x2F, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3F, 0x40, 0x41, 0x43, 0x44, 0x46, 0x48, 0x49, 0x4B, 0x4D, 0x4F, 0x52, 0x55, 0x58, 0x5B, 0x60, 0x67, 0x80, 0x99, 0xA0, 0xA5, 0xA8, 0xAB, 0xAE, 0xB1, 0xB3, 0xB5, 0xB7, 0xB8, 0xBA, 0xBC, 0xBD, 0xBF, 0xC0, 0xC1, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, 0xC8, 0xC9, 0xCA, 0xCB, 0xCC, 0xCD, 0xCE, 0xCF, 0xD0, 0xD1, 0xD1, 0xD2, 0xD3, 0xD4, 0xD5, 0xD5, 0xD6, 0xD7, 0xD8, 0xD8, 0xD9, 0xDA, 0xDA, 0xDB, 0xDC, 0xDC, 0xDD, 0xDE, 0xDE, 0xDF, 0xDF, 0xE0, 0xE1, 0xE1, 0xE2, 0xE2, 0xE3, 0xE3, 0xE4, 0xE5, 0xE5, 0xE6, 0xE6, 0xE7, 0xE7, 0xE8, 0xE8, 0xE9, 0xE9, 0xEA, 0xEA, 0xEB, 0xEB, 0xEC, 0xEC, 0xED, 0xED, 0xED, 0xEE, 0xEE, 0xEF, 0xEF, 0xF0, 0xF0, 0xF1, 0xF1, 0xF1, 0xF2, 0xF2, 0xF3, 0xF3, 0xF3, 0xF4, 0xF4, 0xF5, 0xF5, 0xF5, 0xF6, 0xF6, 0xF7, 0xF7, 0xF7, 0xF8, 0xF8, 0xF9, 0xF9, 0xF9, 0xFA, 0xFA, 0xFA, 0xFB, 0xFB, 0xFC, 0xFC, 0xFC, 0xFD, 0xFD, 0xFD, 0xFE, 0xFE, 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, } // biasAndClamp[x & 1023] is (x + 0x80), clamped to the range [0x00, 0xFF], for // a signed integer x in the range [-512, +511]. var biasAndClamp = [1024]uint8{ 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F, 0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF, 0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7, 0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF, 0xC0, 0xC1, 0xC2, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, 0xC8, 0xC9, 0xCA, 0xCB, 0xCC, 0xCD, 0xCE, 0xCF, 0xD0, 0xD1, 0xD2, 0xD3, 0xD4, 0xD5, 0xD6, 0xD7, 0xD8, 0xD9, 0xDA, 0xDB, 0xDC, 0xDD, 0xDE, 0xDF, 0xE0, 0xE1, 0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7, 0xE8, 0xE9, 0xEA, 0xEB, 0xEC, 0xED, 0xEE, 0xEF, 0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F, 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E, 0x6F, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F, }