Austin Schuh | 3333ec7 | 2022-12-29 16:21:06 -0800 | [diff] [blame^] | 1 | /* Copyright (C) 2013-2016, The Regents of The University of Michigan. |
| 2 | All rights reserved. |
| 3 | This software was developed in the APRIL Robotics Lab under the |
| 4 | direction of Edwin Olson, ebolson@umich.edu. This software may be |
| 5 | available under alternative licensing terms; contact the address above. |
| 6 | Redistribution and use in source and binary forms, with or without |
| 7 | modification, are permitted provided that the following conditions are met: |
| 8 | 1. Redistributions of source code must retain the above copyright notice, this |
| 9 | list of conditions and the following disclaimer. |
| 10 | 2. Redistributions in binary form must reproduce the above copyright notice, |
| 11 | this list of conditions and the following disclaimer in the documentation |
| 12 | and/or other materials provided with the distribution. |
| 13 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND |
| 14 | ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
| 15 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 16 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR |
| 17 | ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
| 18 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 19 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
| 20 | ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 21 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 22 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 23 | The views and conclusions contained in the software and documentation are those |
| 24 | of the authors and should not be interpreted as representing official policies, |
| 25 | either expressed or implied, of the Regents of The University of Michigan. |
| 26 | */ |
| 27 | |
| 28 | #include <stdio.h> |
| 29 | #include <stdlib.h> |
| 30 | #include <assert.h> |
| 31 | #include <stdint.h> |
| 32 | #include <string.h> |
| 33 | |
| 34 | #include "pjpeg.h" |
| 35 | |
| 36 | #include "image_u8.h" |
| 37 | #include "image_u8x3.h" |
| 38 | #include "debug_print.h" |
| 39 | |
| 40 | // https://www.w3.org/Graphics/JPEG/itu-t81.pdf |
| 41 | |
| 42 | void pjpeg_idct_2D_double(int32_t in[64], uint8_t *out, uint32_t outstride); |
| 43 | void pjpeg_idct_2D_u32(int32_t in[64], uint8_t *out, uint32_t outstride); |
| 44 | void pjpeg_idct_2D_nanojpeg(int32_t in[64], uint8_t *out, uint32_t outstride); |
| 45 | |
| 46 | struct pjpeg_huffman_code |
| 47 | { |
| 48 | uint8_t nbits; // how many bits should we actually consume? |
| 49 | uint8_t code; // what is the symbol that was encoded? (not actually a DCT coefficient; see encoding) |
| 50 | }; |
| 51 | |
| 52 | struct pjpeg_decode_state |
| 53 | { |
| 54 | int error; |
| 55 | |
| 56 | uint32_t width, height; |
| 57 | uint8_t *in; |
| 58 | uint32_t inlen; |
| 59 | |
| 60 | uint32_t flags; |
| 61 | |
| 62 | // to decode, we load the next 16 bits of input (generally more |
| 63 | // than we need). We then look up in our code book how many bits |
| 64 | // we have actually consumed. For example, if there was a code |
| 65 | // whose bit sequence was "0", the first 32768 entries would all |
| 66 | // be copies of {.bits=1, .value=XX}; no matter what the following |
| 67 | // 15 bits are, we would get the correct decode. |
| 68 | // |
| 69 | // Can be up to 8 tables; computed as (ACDC * 2 + htidx) |
| 70 | struct pjpeg_huffman_code huff_codes[4][65536]; |
| 71 | int huff_codes_present[4]; |
| 72 | |
| 73 | uint8_t qtab[4][64]; |
| 74 | |
| 75 | int ncomponents; |
| 76 | pjpeg_component_t *components; |
| 77 | |
| 78 | int reset_interval; |
| 79 | int reset_count; |
| 80 | int reset_next; // What reset marker do we expect next? (add 0xd0) |
| 81 | |
| 82 | int debug; |
| 83 | }; |
| 84 | |
| 85 | // from K.3.3.1 (page 158) |
| 86 | static uint8_t mjpeg_dht[] = { // header |
| 87 | 0xFF,0xC4,0x01,0xA2, |
| 88 | |
| 89 | ///////////////////////////////////////////////////////////// |
| 90 | // luminance dc coefficients. |
| 91 | // DC table 0 |
| 92 | 0x00, |
| 93 | // code lengths |
| 94 | 0x00,0x01,0x05,0x01,0x01,0x01,0x01,0x01,0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, |
| 95 | // values |
| 96 | 0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0A,0x0B, |
| 97 | |
| 98 | ///////////////////////////////////////////////////////////// |
| 99 | // chrominance DC coefficents |
| 100 | // DC table 1 |
| 101 | 0x01, |
| 102 | // code lengths |
| 103 | 0x00,0x03,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x00,0x00,0x00,0x00,0x00, |
| 104 | // values |
| 105 | 0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0A,0x0B, |
| 106 | |
| 107 | ///////////////////////////////////////////////////////////// |
| 108 | // luminance AC coefficients |
| 109 | // AC table 0 |
| 110 | 0x10, |
| 111 | // code lengths |
| 112 | 0x00,0x02,0x01,0x03,0x03,0x02,0x04,0x03,0x05,0x05,0x04,0x04,0x00,0x00,0x01,0x7D, |
| 113 | // codes |
| 114 | 0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61, |
| 115 | 0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xA1,0x08,0x23,0x42,0xB1,0xC1,0x15,0x52,0xD1,0xF0,0x24, |
| 116 | 0x33,0x62,0x72,0x82,0x09,0x0A,0x16,0x17,0x18,0x19,0x1A,0x25,0x26,0x27,0x28,0x29,0x2A,0x34, |
| 117 | 0x35,0x36,0x37,0x38,0x39,0x3A,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4A,0x53,0x54,0x55,0x56, |
| 118 | 0x57,0x58,0x59,0x5A,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6A,0x73,0x74,0x75,0x76,0x77,0x78, |
| 119 | 0x79,0x7A,0x83,0x84,0x85,0x86,0x87,0x88,0x89,0x8A,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99, |
| 120 | 0x9A,0xA2,0xA3,0xA4,0xA5,0xA6,0xA7,0xA8,0xA9,0xAA,0xB2,0xB3,0xB4,0xB5,0xB6,0xB7,0xB8,0xB9, |
| 121 | 0xBA,0xC2,0xC3,0xC4,0xC5,0xC6,0xC7,0xC8,0xC9,0xCA,0xD2,0xD3,0xD4,0xD5,0xD6,0xD7,0xD8,0xD9, |
| 122 | 0xDA,0xE1,0xE2,0xE3,0xE4,0xE5,0xE6,0xE7,0xE8,0xE9,0xEA,0xF1,0xF2,0xF3,0xF4,0xF5,0xF6,0xF7, |
| 123 | 0xF8,0xF9,0xFA, |
| 124 | |
| 125 | ///////////////////////////////////////////////////////////// |
| 126 | // chrominance DC coefficients |
| 127 | // DC table 1 |
| 128 | 0x11, |
| 129 | // code lengths |
| 130 | 0x00,0x02,0x01,0x02,0x04,0x04,0x03,0x04,0x07,0x05,0x04,0x04,0x00,0x01,0x02,0x77, |
| 131 | // values |
| 132 | 0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71, |
| 133 | 0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91,0xA1,0xB1,0xC1,0x09,0x23,0x33,0x52,0xF0,0x15,0x62, |
| 134 | 0x72,0xD1,0x0A,0x16,0x24,0x34,0xE1,0x25,0xF1,0x17,0x18,0x19,0x1A,0x26,0x27,0x28,0x29,0x2A, |
| 135 | 0x35,0x36,0x37,0x38,0x39,0x3A,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4A,0x53,0x54,0x55,0x56, |
| 136 | 0x57,0x58,0x59,0x5A,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6A,0x73,0x74,0x75,0x76,0x77,0x78, |
| 137 | 0x79,0x7A,0x82,0x83,0x84,0x85,0x86,0x87,0x88,0x89,0x8A,0x92,0x93,0x94,0x95,0x96,0x97,0x98, |
| 138 | 0x99,0x9A,0xA2,0xA3,0xA4,0xA5,0xA6,0xA7,0xA8,0xA9,0xAA,0xB2,0xB3,0xB4,0xB5,0xB6,0xB7,0xB8, |
| 139 | 0xB9,0xBA,0xC2,0xC3,0xC4,0xC5,0xC6,0xC7,0xC8,0xC9,0xCA,0xD2,0xD3,0xD4,0xD5,0xD6,0xD7,0xD8, |
| 140 | 0xD9,0xDA,0xE2,0xE3,0xE4,0xE5,0xE6,0xE7,0xE8,0xE9,0xEA,0xF2,0xF3,0xF4,0xF5,0xF6,0xF7,0xF8, |
| 141 | 0xF9,0xFA |
| 142 | }; |
| 143 | |
| 144 | static inline uint8_t max_u8(uint8_t a, uint8_t b) |
| 145 | { |
| 146 | return a > b ? a : b; |
| 147 | } |
| 148 | |
| 149 | // order of coefficients in each DC block |
| 150 | static const char ZZ[64] = { 0, 1, 8, 16, 9, 2, 3, 10, |
| 151 | 17, 24, 32, 25, 18, 11, 4, 5, |
| 152 | 12, 19, 26, 33, 40, 48, 41, 34, |
| 153 | 27, 20, 13, 6, 7, 14, 21, 28, |
| 154 | 35, 42, 49, 56, 57, 50, 43, 36, |
| 155 | 29, 22, 15, 23, 30, 37, 44, 51, |
| 156 | 58, 59, 52, 45, 38, 31, 39, 46, |
| 157 | 53, 60, 61, 54, 47, 55, 62, 63 }; |
| 158 | |
| 159 | |
| 160 | |
| 161 | struct bit_decoder |
| 162 | { |
| 163 | uint8_t *in; |
| 164 | uint32_t inpos; |
| 165 | uint32_t inlen; |
| 166 | |
| 167 | uint32_t bits; // the low order bits contain the next nbits_avail bits. |
| 168 | |
| 169 | int nbits_avail; // how many bits in 'bits' (left aligned) are valid? |
| 170 | |
| 171 | int error; |
| 172 | }; |
| 173 | |
| 174 | // ensure that at least 'nbits' of data is available in the bit decoder. |
| 175 | static inline void bd_ensure(struct bit_decoder *bd, int nbits) |
| 176 | { |
| 177 | while (bd->nbits_avail < nbits) { |
| 178 | |
| 179 | if (bd->inpos >= bd->inlen) { |
| 180 | printf("hallucinating 1s!\n"); |
| 181 | // we hit end of stream hallucinate an infinite stream of 1s |
| 182 | bd->bits = (bd->bits << 8) | 0xff; |
| 183 | bd->nbits_avail += 8; |
| 184 | continue; |
| 185 | } |
| 186 | |
| 187 | uint8_t nextbyte = bd->in[bd->inpos]; |
| 188 | bd->inpos++; |
| 189 | |
| 190 | if (nextbyte == 0xff && bd->inpos < bd->inlen && bd->in[bd->inpos] == 0x00) { |
| 191 | // a stuffed byte |
| 192 | nextbyte = 0xff; |
| 193 | bd->inpos++; |
| 194 | } |
| 195 | |
| 196 | // it's an ordinary byte |
| 197 | bd->bits = (bd->bits << 8) | nextbyte; |
| 198 | bd->nbits_avail += 8; |
| 199 | } |
| 200 | } |
| 201 | |
| 202 | static inline uint32_t bd_peek_bits(struct bit_decoder *bd, int nbits) |
| 203 | { |
| 204 | bd_ensure(bd, nbits); |
| 205 | |
| 206 | return (bd->bits >> (bd->nbits_avail - nbits)) & ((1 << nbits) - 1); |
| 207 | } |
| 208 | |
| 209 | static inline uint32_t bd_consume_bits(struct bit_decoder *bd, int nbits) |
| 210 | { |
| 211 | assert(nbits < 32); |
| 212 | |
| 213 | bd_ensure(bd, nbits); |
| 214 | |
| 215 | uint32_t v = (bd->bits >> (bd->nbits_avail - nbits)) & ((1 << nbits) - 1); |
| 216 | |
| 217 | bd->nbits_avail -= nbits; |
| 218 | |
| 219 | return v; |
| 220 | } |
| 221 | |
| 222 | // discard without regard for byte stuffing! |
| 223 | static inline void bd_discard_bytes(struct bit_decoder *bd, int nbytes) |
| 224 | { |
| 225 | assert(bd->nbits_avail == 0); |
| 226 | bd->inpos += nbytes; |
| 227 | } |
| 228 | |
| 229 | static inline int bd_has_more(struct bit_decoder *bd) |
| 230 | { |
| 231 | return bd->nbits_avail > 0 || bd->inpos < bd->inlen; |
| 232 | } |
| 233 | |
| 234 | // throw away up to 7 bits of data so that the next data returned |
| 235 | // began on a byte boundary. |
| 236 | static inline void bd_discard_to_byte_boundary(struct bit_decoder *bd) |
| 237 | { |
| 238 | bd->nbits_avail -= (bd->nbits_avail & 7); |
| 239 | } |
| 240 | |
| 241 | static inline uint32_t bd_get_offset(struct bit_decoder *bd) |
| 242 | { |
| 243 | return bd->inpos - bd->nbits_avail / 8; |
| 244 | } |
| 245 | |
| 246 | static int pjpeg_decode_buffer(struct pjpeg_decode_state *pjd) |
| 247 | { |
| 248 | // XXX TODO Include sanity check that this is actually a JPG |
| 249 | |
| 250 | struct bit_decoder bd; |
| 251 | memset(&bd, 0, sizeof(struct bit_decoder)); |
| 252 | bd.in = pjd->in; |
| 253 | bd.inpos = 0; |
| 254 | bd.inlen = pjd->inlen; |
| 255 | |
| 256 | int marker_sync_skipped = 0; |
| 257 | int marker_sync_skipped_from_offset = 0; |
| 258 | |
| 259 | while (bd_has_more(&bd)) { |
| 260 | |
| 261 | uint32_t marker_offset = bd_get_offset(&bd); |
| 262 | |
| 263 | // Look for the 0xff that signifies the beginning of a marker |
| 264 | bd_discard_to_byte_boundary(&bd); |
| 265 | |
| 266 | while (bd_consume_bits(&bd, 8) != 0xff) { |
| 267 | if (marker_sync_skipped == 0) |
| 268 | marker_sync_skipped_from_offset = marker_offset; |
| 269 | marker_sync_skipped++; |
| 270 | continue; |
| 271 | } |
| 272 | |
| 273 | if (marker_sync_skipped) { |
| 274 | printf("%08x: skipped %04x bytes\n", marker_sync_skipped_from_offset, marker_sync_skipped); |
| 275 | marker_sync_skipped = 0; |
| 276 | } |
| 277 | |
| 278 | uint8_t marker = bd_consume_bits(&bd, 8); |
| 279 | |
| 280 | // printf("marker %08x : %02x\n", marker_offset, marker); |
| 281 | |
| 282 | switch (marker) { |
| 283 | |
| 284 | case 0xd8: // start of image. Great, continue. |
| 285 | continue; |
| 286 | |
| 287 | // below are the markers that A) we don't care about |
| 288 | // that B) encode length as two bytes. |
| 289 | // |
| 290 | // Note: Other unknown fields should not be added since |
| 291 | // we should be able to skip over them by looking for |
| 292 | // the next marker byte. |
| 293 | case 0xe0: // JFIF header. |
| 294 | case 0xe1: // EXIF header (Yuck: Payload may contain 0xff 0xff!) |
| 295 | case 0xe2: // ICC Profile. (Yuck: payload may contain 0xff 0xff!) |
| 296 | case 0xe6: // some other common header |
| 297 | case 0xfe: // Comment |
| 298 | { |
| 299 | uint16_t length = bd_consume_bits(&bd, 16); |
| 300 | bd_discard_bytes(&bd, length - 2); |
| 301 | continue; |
| 302 | } |
| 303 | |
| 304 | case 0xdb: { // DQT Define Quantization Table |
| 305 | uint16_t length = bd_consume_bits(&bd, 16); |
| 306 | |
| 307 | if (((length-2) % 65) != 0) |
| 308 | return PJPEG_ERR_DQT; |
| 309 | |
| 310 | // can contain multiple DQTs |
| 311 | for (int offset = 0; offset < length - 2; offset += 65) { |
| 312 | |
| 313 | // pq: quant table element precision. 0=8bit, 1=16bit. |
| 314 | // tq: quant table destination id. |
| 315 | uint8_t pqtq = bd_consume_bits(&bd, 8); |
| 316 | |
| 317 | if ((pqtq & 0xf0) != 0 || (pqtq & 0x0f) >= 4) |
| 318 | return PJPEG_ERR_DQT; |
| 319 | |
| 320 | uint8_t id = pqtq & 3; |
| 321 | |
| 322 | for (int i = 0; i < 64; i++) |
| 323 | pjd->qtab[id][i] = bd_consume_bits(&bd, 8); |
| 324 | } |
| 325 | |
| 326 | break; |
| 327 | } |
| 328 | |
| 329 | case 0xc0: { // SOF, non-differential, huffman, baseline |
| 330 | uint16_t length = bd_consume_bits(&bd, 16); |
| 331 | (void) length; |
| 332 | |
| 333 | uint8_t p = bd_consume_bits(&bd, 8); // precision |
| 334 | if (p != 8) |
| 335 | return PJPEG_ERR_SOF; |
| 336 | |
| 337 | pjd->height = bd_consume_bits(&bd, 16); |
| 338 | pjd->width = bd_consume_bits(&bd, 16); |
| 339 | |
| 340 | // printf("%d x %d\n", pjd->height, pjd->width); |
| 341 | |
| 342 | int nf = bd_consume_bits(&bd, 8); // # image components |
| 343 | |
| 344 | if (nf < 1 || nf > 3) |
| 345 | return PJPEG_ERR_SOF; |
| 346 | |
| 347 | pjd->ncomponents = nf; |
| 348 | pjd->components = calloc(nf, sizeof(struct pjpeg_component)); |
| 349 | |
| 350 | for (int i = 0; i < nf; i++) { |
| 351 | // comp. identifier |
| 352 | pjd->components[i].id = bd_consume_bits(&bd, 8); |
| 353 | |
| 354 | // horiz/vert sampling |
| 355 | pjd->components[i].hv = bd_consume_bits(&bd, 8); |
| 356 | pjd->components[i].scaley = pjd->components[i].hv & 0x0f; |
| 357 | pjd->components[i].scalex = pjd->components[i].hv >> 4; |
| 358 | |
| 359 | // which quant table? |
| 360 | pjd->components[i].tq = bd_consume_bits(&bd, 8); |
| 361 | } |
| 362 | break; |
| 363 | } |
| 364 | |
| 365 | case 0xc1: // SOF, non-differential, huffman, extended DCT |
| 366 | case 0xc2: // SOF, non-differential, huffman, progressive DCT |
| 367 | case 0xc3: // SOF, non-differential, huffman, lossless |
| 368 | case 0xc5: // SOF, differential, huffman, baseline DCT |
| 369 | case 0xc6: // SOF, differential, huffman, progressive |
| 370 | case 0xc7: // SOF, differential, huffman, lossless |
| 371 | case 0xc8: // reserved |
| 372 | case 0xc9: // SOF, non-differential, arithmetic, extended |
| 373 | case 0xca: // SOF, non-differential, arithmetic, progressive |
| 374 | case 0xcb: // SOF, non-differential, arithmetic, lossless |
| 375 | case 0xcd: // SOF, differential, arithmetic, sequential |
| 376 | case 0xce: // SOF, differential, arithmetic, progressive |
| 377 | case 0xcf: // SOF, differential, arithmetic, lossless |
| 378 | { |
| 379 | printf("pjepg.c: unsupported JPEG type %02x\n", marker); |
| 380 | return PJEPG_ERR_UNSUPPORTED; |
| 381 | } |
| 382 | |
| 383 | case 0xc4: { // DHT Define Huffman Tables |
| 384 | // [ED: the encoding of these tables is really quite |
| 385 | // clever!] |
| 386 | uint16_t length = bd_consume_bits(&bd, 16); |
| 387 | length = length - 2; |
| 388 | |
| 389 | while (length > 0) { |
| 390 | uint8_t TcTh = bd_consume_bits(&bd, 8); |
| 391 | length--; |
| 392 | uint8_t Tc = (TcTh >> 4); |
| 393 | int Th = TcTh & 0x0f; // which index are we using? |
| 394 | |
| 395 | if (Tc >= 2 || Th >= 2) |
| 396 | // Tc must be either AC=1 or DC=0. |
| 397 | // Th must be less than 2 |
| 398 | return PJPEG_ERR_DHT; |
| 399 | |
| 400 | int htidx = Tc*2 + Th; |
| 401 | |
| 402 | uint8_t L[17]; // how many symbols of each bit length? |
| 403 | L[0] = 0; // no 0 bit codes :) |
| 404 | for (int nbits = 1; nbits <= 16; nbits++) { |
| 405 | L[nbits] = bd_consume_bits(&bd, 8); |
| 406 | length -= L[nbits]; |
| 407 | } |
| 408 | length -= 16; |
| 409 | |
| 410 | uint32_t code_pos = 0; |
| 411 | |
| 412 | for (int nbits = 1; nbits <= 16; nbits++) { |
| 413 | int nvalues = L[nbits]; |
| 414 | |
| 415 | // how many entries will we fill? |
| 416 | // (a 1 bit code will fill 32768, a 2 bit code 16384, ...) |
| 417 | uint32_t ncodes = (1 << (16 - nbits)); |
| 418 | |
| 419 | // consume the values... |
| 420 | for (int vi = 0; vi < nvalues; vi++) { |
| 421 | uint8_t code = bd_consume_bits(&bd, 8); |
| 422 | |
| 423 | if (code_pos + ncodes > 0xffff) |
| 424 | return PJPEG_ERR_DHT; |
| 425 | |
| 426 | for (int ci = 0; ci < ncodes; ci++) { |
| 427 | pjd->huff_codes[htidx][code_pos].nbits = nbits; |
| 428 | pjd->huff_codes[htidx][code_pos].code = code; |
| 429 | code_pos++; |
| 430 | } |
| 431 | } |
| 432 | } |
| 433 | pjd->huff_codes_present[htidx] = 1; |
| 434 | } |
| 435 | break; |
| 436 | } |
| 437 | |
| 438 | // a sequentially-encoded JPG has one SOS segment. A |
| 439 | // progressive JPG will have multiple SOS segments. |
| 440 | case 0xda: { // Start Of Scan (SOS) |
| 441 | |
| 442 | // Note that this marker frame (and its encoded |
| 443 | // length) does NOT include the bitstream that |
| 444 | // follows. |
| 445 | |
| 446 | uint16_t length = bd_consume_bits(&bd, 16); |
| 447 | (void) length; |
| 448 | |
| 449 | // number of components in this scan |
| 450 | uint8_t ns = bd_consume_bits(&bd, 8); |
| 451 | |
| 452 | // for each component, what is the index into our pjd->components[] array? |
| 453 | uint8_t *comp_idx = calloc(ns, sizeof(uint8_t)); |
| 454 | |
| 455 | for (int i = 0; i < ns; i++) { |
| 456 | // component name |
| 457 | uint8_t cs = bd_consume_bits(&bd, 8); |
| 458 | |
| 459 | int found = 0; |
| 460 | for (int j = 0; j < pjd->ncomponents; j++) { |
| 461 | |
| 462 | if (cs == pjd->components[j].id) { |
| 463 | // which huff tables will we use for |
| 464 | // DC (high 4 bits) and AC (low 4 bits) |
| 465 | pjd->components[j].tda = bd_consume_bits(&bd, 8); |
| 466 | comp_idx[i] = j; |
| 467 | found = 1; |
| 468 | break; |
| 469 | } |
| 470 | } |
| 471 | |
| 472 | if (!found) |
| 473 | return PJPEG_ERR_SOS; |
| 474 | } |
| 475 | |
| 476 | // start of spectral selection. baseline == 0 |
| 477 | uint8_t ss = bd_consume_bits(&bd, 8); |
| 478 | |
| 479 | // end of spectral selection. baseline == 0x3f |
| 480 | uint8_t se = bd_consume_bits(&bd, 8); |
| 481 | |
| 482 | // successive approximation bits. baseline == 0 |
| 483 | uint8_t Ahl = bd_consume_bits(&bd, 8); |
| 484 | |
| 485 | if (ss != 0 || se != 0x3f || Ahl != 0x00) |
| 486 | return PJPEG_ERR_SOS; |
| 487 | |
| 488 | // compute the dimensions of each MCU in pixels |
| 489 | int maxmcux = 0, maxmcuy = 0; |
| 490 | for (int i = 0; i < ns; i++) { |
| 491 | struct pjpeg_component *comp = &pjd->components[comp_idx[i]]; |
| 492 | |
| 493 | maxmcux = max_u8(maxmcux, comp->scalex * 8); |
| 494 | maxmcuy = max_u8(maxmcuy, comp->scaley * 8); |
| 495 | } |
| 496 | |
| 497 | // how many MCU blocks are required to encode the whole image? |
| 498 | int mcus_x = (pjd->width + maxmcux - 1) / maxmcux; |
| 499 | int mcus_y = (pjd->height + maxmcuy - 1) / maxmcuy; |
| 500 | |
| 501 | if (0) |
| 502 | printf("Image has %d x %d MCU blocks, each %d x %d pixels\n", |
| 503 | mcus_x, mcus_y, maxmcux, maxmcuy); |
| 504 | |
| 505 | // allocate output storage |
| 506 | for (int i = 0; i < ns; i++) { |
| 507 | struct pjpeg_component *comp = &pjd->components[comp_idx[i]]; |
| 508 | comp->width = mcus_x * comp->scalex * 8; |
| 509 | comp->height = mcus_y * comp->scaley * 8; |
| 510 | comp->stride = comp->width; |
| 511 | |
| 512 | int alignment = 32; |
| 513 | if ((comp->stride % alignment) != 0) |
| 514 | comp->stride += alignment - (comp->stride % alignment); |
| 515 | |
| 516 | comp->data = calloc(comp->height * comp->stride, 1); |
| 517 | } |
| 518 | |
| 519 | |
| 520 | // each component has its own DC prediction |
| 521 | int32_t *dcpred = calloc(ns, sizeof(int32_t)); |
| 522 | |
| 523 | pjd->reset_count = 0; |
| 524 | |
| 525 | for (int mcu_y = 0; mcu_y < mcus_y; mcu_y++) { |
| 526 | for (int mcu_x = 0; mcu_x < mcus_x; mcu_x++) { |
| 527 | |
| 528 | // the next two bytes in the input stream |
| 529 | // should be 0xff 0xdN, where N is the next |
| 530 | // reset counter. |
| 531 | // |
| 532 | // Our bit decoder may have already shifted |
| 533 | // these into the buffer. Consequently, we |
| 534 | // want to use our bit decoding functions to |
| 535 | // check for the marker. But we must first |
| 536 | // discard any fractional bits left. |
| 537 | if (pjd->reset_interval > 0 && pjd->reset_count == pjd->reset_interval) { |
| 538 | |
| 539 | // RST markers are byte-aligned, so force |
| 540 | // the bit-decoder to the next byte |
| 541 | // boundary. |
| 542 | bd_discard_to_byte_boundary(&bd); |
| 543 | |
| 544 | while (1) { |
| 545 | int32_t value = bd_consume_bits(&bd, 8); |
| 546 | if (bd.inpos > bd.inlen) |
| 547 | return PJPEG_ERR_EOF; |
| 548 | if (value == 0xff) |
| 549 | break; |
| 550 | printf("RST SYNC\n"); |
| 551 | } |
| 552 | |
| 553 | int32_t marker_32 = bd_consume_bits(&bd, 8); |
| 554 | |
| 555 | // printf("%04x: RESET? %02x\n", *bd.inpos, marker_32); |
| 556 | if (marker_32 != (0xd0 + pjd->reset_next)) |
| 557 | return PJPEG_ERR_RESET; |
| 558 | |
| 559 | pjd->reset_count = 0; |
| 560 | pjd->reset_next = (pjd->reset_next + 1) & 0x7; |
| 561 | |
| 562 | memset(dcpred, 0, sizeof(*dcpred)); |
| 563 | } |
| 564 | |
| 565 | for (int nsidx = 0; nsidx < ns; nsidx++) { |
| 566 | |
| 567 | struct pjpeg_component *comp = &pjd->components[comp_idx[nsidx]]; |
| 568 | |
| 569 | int32_t block[64]; |
| 570 | |
| 571 | int qtabidx = comp->tq; // which quant table? |
| 572 | |
| 573 | for (int sby = 0; sby < comp->scaley; sby++) { |
| 574 | for (int sbx = 0; sbx < comp->scalex; sbx++) { |
| 575 | // decode block for component nsidx |
| 576 | memset(block, 0, sizeof(block)); |
| 577 | |
| 578 | int dc_huff_table_idx = comp->tda >> 4; |
| 579 | int ac_huff_table_idx = 2 + (comp->tda & 0x0f); |
| 580 | |
| 581 | if (!pjd->huff_codes_present[dc_huff_table_idx] || |
| 582 | !pjd->huff_codes_present[ac_huff_table_idx]) |
| 583 | return PJPEG_ERR_MISSING_DHT; // probably an MJPEG. |
| 584 | |
| 585 | |
| 586 | if (1) { |
| 587 | // do DC coefficient |
| 588 | uint32_t next16 = bd_peek_bits(&bd, 16); |
| 589 | struct pjpeg_huffman_code *huff_code = &pjd->huff_codes[dc_huff_table_idx][next16]; |
| 590 | bd_consume_bits(&bd, huff_code->nbits); |
| 591 | |
| 592 | int ssss = huff_code->code & 0x0f; // ssss == number of additional bits to read |
| 593 | int32_t value = bd_consume_bits(&bd, ssss); |
| 594 | |
| 595 | // if high bit is clear, it's negative |
| 596 | if ((value & (1 << (ssss-1))) == 0) |
| 597 | value += ((-1) << ssss) + 1; |
| 598 | |
| 599 | dcpred[nsidx] += value; |
| 600 | block[0] = dcpred[nsidx] * pjd->qtab[qtabidx][0]; |
| 601 | } |
| 602 | |
| 603 | if (1) { |
| 604 | // do AC coefficients |
| 605 | for (int coeff = 1; coeff < 64; coeff++) { |
| 606 | |
| 607 | uint32_t next16 = bd_peek_bits(&bd, 16); |
| 608 | |
| 609 | struct pjpeg_huffman_code *huff_code = &pjd->huff_codes[ac_huff_table_idx][next16]; |
| 610 | bd_consume_bits(&bd, huff_code->nbits); |
| 611 | |
| 612 | if (huff_code->code == 0) { |
| 613 | break; // EOB |
| 614 | } |
| 615 | |
| 616 | int rrrr = huff_code->code >> 4; // run length of zeros |
| 617 | int ssss = huff_code->code & 0x0f; |
| 618 | |
| 619 | int32_t value = bd_consume_bits(&bd, ssss); |
| 620 | |
| 621 | // if high bit is clear, it's negative |
| 622 | if ((value & (1 << (ssss-1))) == 0) |
| 623 | value += ((-1) << ssss) + 1; |
| 624 | |
| 625 | coeff += rrrr; |
| 626 | |
| 627 | block[(int) ZZ[coeff]] = value * pjd->qtab[qtabidx][coeff]; |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | // do IDCT |
| 632 | |
| 633 | // output block's upper-left |
| 634 | // coordinate (in pixels) is |
| 635 | // (comp_x, comp_y). |
| 636 | uint32_t comp_x = (mcu_x * comp->scalex + sbx) * 8; |
| 637 | uint32_t comp_y = (mcu_y * comp->scaley + sby) * 8; |
| 638 | uint32_t dataidx = comp_y * comp->stride + comp_x; |
| 639 | |
| 640 | // pjpeg_idct_2D_u32(block, &comp->data[dataidx], comp->stride); |
| 641 | pjpeg_idct_2D_nanojpeg(block, &comp->data[dataidx], comp->stride); |
| 642 | } |
| 643 | } |
| 644 | } |
| 645 | |
| 646 | pjd->reset_count++; |
| 647 | // printf("%04x: reset count %d / %d\n", pjd->inpos, pjd->reset_count, pjd->reset_interval); |
| 648 | |
| 649 | } |
| 650 | } |
| 651 | |
| 652 | free(dcpred); |
| 653 | free(comp_idx); |
| 654 | |
| 655 | break; |
| 656 | } |
| 657 | |
| 658 | case 0xd9: { // EOI End of Image |
| 659 | goto got_end_of_image; |
| 660 | } |
| 661 | |
| 662 | case 0xdd: { // Define Restart Interval |
| 663 | uint16_t length = bd_consume_bits(&bd, 16); |
| 664 | if (length != 4) |
| 665 | return PJPEG_ERR_DRI; |
| 666 | |
| 667 | // reset interval measured in the number of MCUs |
| 668 | pjd->reset_interval = bd_consume_bits(&bd, 16); |
| 669 | |
| 670 | break; |
| 671 | } |
| 672 | |
| 673 | default: { |
| 674 | printf("pjepg: Unknown marker %02x at offset %04x\n", marker, marker_offset); |
| 675 | |
| 676 | // try to skip it. |
| 677 | uint16_t length = bd_consume_bits(&bd, 16); |
| 678 | bd_discard_bytes(&bd, length - 2); |
| 679 | continue; |
| 680 | } |
| 681 | } // switch (marker) |
| 682 | } // while inpos < inlen |
| 683 | |
| 684 | got_end_of_image: |
| 685 | |
| 686 | return PJPEG_OKAY; |
| 687 | } |
| 688 | |
| 689 | void pjpeg_destroy(pjpeg_t *pj) |
| 690 | { |
| 691 | if (!pj) |
| 692 | return; |
| 693 | |
| 694 | for (int i = 0; i < pj->ncomponents; i++) |
| 695 | free(pj->components[i].data); |
| 696 | free(pj->components); |
| 697 | |
| 698 | free(pj); |
| 699 | } |
| 700 | |
| 701 | |
| 702 | // just grab the first component. |
| 703 | image_u8_t *pjpeg_to_u8_baseline(pjpeg_t *pj) |
| 704 | { |
| 705 | assert(pj->ncomponents > 0); |
| 706 | |
| 707 | pjpeg_component_t *comp = &pj->components[0]; |
| 708 | |
| 709 | assert(comp->width >= pj->width && comp->height >= pj->height); |
| 710 | |
| 711 | image_u8_t *im = image_u8_create(pj->width, pj->height); |
| 712 | for (int y = 0; y < im->height; y++) |
| 713 | memcpy(&im->buf[y*im->stride], &comp->data[y*comp->stride], pj->width); |
| 714 | |
| 715 | return im; |
| 716 | } |
| 717 | |
| 718 | static inline uint8_t clampd(double v) |
| 719 | { |
| 720 | if (v < 0) |
| 721 | return 0; |
| 722 | if (v > 255) |
| 723 | return 255; |
| 724 | |
| 725 | return (uint8_t) v; |
| 726 | } |
| 727 | |
| 728 | static inline uint8_t clamp_u8(int32_t v) |
| 729 | { |
| 730 | if (v < 0) |
| 731 | return 0; |
| 732 | if (v > 255) |
| 733 | return 255; |
| 734 | return v; |
| 735 | } |
| 736 | |
| 737 | // color conversion formulas taken from JFIF spec v 1.02 |
| 738 | image_u8x3_t *pjpeg_to_u8x3_baseline(pjpeg_t *pj) |
| 739 | { |
| 740 | assert(pj->ncomponents == 3); |
| 741 | |
| 742 | pjpeg_component_t *Y = &pj->components[0]; |
| 743 | pjpeg_component_t *Cb = &pj->components[1]; |
| 744 | pjpeg_component_t *Cr = &pj->components[2]; |
| 745 | |
| 746 | int Cb_factor_y = Y->height / Cb->height; |
| 747 | int Cb_factor_x = Y->width / Cb->width; |
| 748 | |
| 749 | int Cr_factor_y = Y->height / Cr->height; |
| 750 | int Cr_factor_x = Y->width / Cr->width; |
| 751 | |
| 752 | image_u8x3_t *im = image_u8x3_create(pj->width, pj->height); |
| 753 | |
| 754 | if (Cr_factor_y == 1 && Cr_factor_x == 1 && Cb_factor_y == 1 && Cb_factor_x == 1) { |
| 755 | |
| 756 | for (int y = 0; y < pj->height; y++) { |
| 757 | for (int x = 0; x < pj->width; x++) { |
| 758 | int32_t y_val = Y->data[y*Y->stride + x] * 65536; |
| 759 | int32_t cb_val = Cb->data[y*Cb->stride + x] - 128; |
| 760 | int32_t cr_val = Cr->data[y*Cr->stride + x] - 128; |
| 761 | |
| 762 | int32_t r_val = y_val + 91881 * cr_val; |
| 763 | int32_t g_val = y_val + -22554 * cb_val - 46802 * cr_val; |
| 764 | int32_t b_val = y_val + 116130 * cb_val; |
| 765 | |
| 766 | im->buf[y*im->stride + 3*x + 0 ] = clamp_u8(r_val >> 16); |
| 767 | im->buf[y*im->stride + 3*x + 1 ] = clamp_u8(g_val >> 16); |
| 768 | im->buf[y*im->stride + 3*x + 2 ] = clamp_u8(b_val >> 16); |
| 769 | } |
| 770 | } |
| 771 | } else if (Cb_factor_y == Cr_factor_y && Cb_factor_x == Cr_factor_x) { |
| 772 | for (int by = 0; by < pj->height / Cb_factor_y; by++) { |
| 773 | for (int bx = 0; bx < pj->width / Cb_factor_x; bx++) { |
| 774 | |
| 775 | int32_t cb_val = Cb->data[by*Cb->stride + bx] - 128; |
| 776 | int32_t cr_val = Cr->data[by*Cr->stride + bx] - 128; |
| 777 | |
| 778 | int32_t r0 = 91881 * cr_val; |
| 779 | int32_t g0 = -22554 * cb_val - 46802 * cr_val; |
| 780 | int32_t b0 = 116130 * cb_val; |
| 781 | |
| 782 | for (int dy = 0; dy < Cb_factor_y; dy++) { |
| 783 | int y = by*Cb_factor_y + dy; |
| 784 | |
| 785 | for (int dx = 0; dx < Cb_factor_x; dx++) { |
| 786 | int x = bx*Cb_factor_x + dx; |
| 787 | |
| 788 | int32_t y_val = Y->data[y*Y->stride + x] * 65536; |
| 789 | |
| 790 | int32_t r_val = r0 + y_val; |
| 791 | int32_t g_val = g0 + y_val; |
| 792 | int32_t b_val = b0 + y_val; |
| 793 | |
| 794 | im->buf[y*im->stride + 3*x + 0 ] = clamp_u8(r_val >> 16); |
| 795 | im->buf[y*im->stride + 3*x + 1 ] = clamp_u8(g_val >> 16); |
| 796 | im->buf[y*im->stride + 3*x + 2 ] = clamp_u8(b_val >> 16); |
| 797 | } |
| 798 | } |
| 799 | } |
| 800 | } |
| 801 | } else { |
| 802 | |
| 803 | for (int y = 0; y < pj->height; y++) { |
| 804 | for (int x = 0; x < pj->width; x++) { |
| 805 | int32_t y_val = Y->data[y*Y->stride + x]; |
| 806 | int32_t cb_val = Cb->data[(y / Cb_factor_y)*Cb->stride + (x / Cb_factor_x)] - 128; |
| 807 | int32_t cr_val = Cr->data[(y / Cr_factor_y)*Cr->stride + (x / Cr_factor_x)] - 128; |
| 808 | |
| 809 | uint8_t r_val = clampd(y_val + 1.402 * cr_val); |
| 810 | uint8_t g_val = clampd(y_val - 0.34414 * cb_val - 0.71414 * cr_val); |
| 811 | uint8_t b_val = clampd(y_val + 1.772 * cb_val); |
| 812 | |
| 813 | im->buf[y*im->stride + 3*x + 0 ] = r_val; |
| 814 | im->buf[y*im->stride + 3*x + 1 ] = g_val; |
| 815 | im->buf[y*im->stride + 3*x + 2 ] = b_val; |
| 816 | } |
| 817 | } |
| 818 | } |
| 819 | |
| 820 | return im; |
| 821 | } |
| 822 | |
| 823 | /////////////////////////////////////////////////////////////////// |
| 824 | // returns NULL if file loading fails. |
| 825 | pjpeg_t *pjpeg_create_from_file(const char *path, uint32_t flags, int *error) |
| 826 | { |
| 827 | FILE *f = fopen(path, "rb"); |
| 828 | if (f == NULL) |
| 829 | return NULL; |
| 830 | |
| 831 | fseek(f, 0, SEEK_END); |
| 832 | long buflen = ftell(f); |
| 833 | |
| 834 | uint8_t *buf = malloc(buflen); |
| 835 | fseek(f, 0, SEEK_SET); |
| 836 | int res = fread(buf, 1, buflen, f); |
| 837 | |
| 838 | if ( ferror(f) ){ |
| 839 | debug_print ("Read failed"); |
| 840 | clearerr(f); |
| 841 | } |
| 842 | |
| 843 | fclose(f); |
| 844 | if (res != buflen) { |
| 845 | free(buf); |
| 846 | if (error) |
| 847 | *error = PJPEG_ERR_FILE; |
| 848 | return NULL; |
| 849 | } |
| 850 | |
| 851 | pjpeg_t *pj = pjpeg_create_from_buffer(buf, buflen, flags, error); |
| 852 | |
| 853 | free(buf); |
| 854 | return pj; |
| 855 | } |
| 856 | |
| 857 | pjpeg_t *pjpeg_create_from_buffer(uint8_t *buf, int buflen, uint32_t flags, int *error) |
| 858 | { |
| 859 | struct pjpeg_decode_state pjd; |
| 860 | memset(&pjd, 0, sizeof(pjd)); |
| 861 | |
| 862 | if (flags & PJPEG_MJPEG) { |
| 863 | pjd.in = mjpeg_dht; |
| 864 | pjd.inlen = sizeof(mjpeg_dht); |
| 865 | int result = pjpeg_decode_buffer(&pjd); |
| 866 | assert(result == 0); |
| 867 | } |
| 868 | |
| 869 | pjd.in = buf; |
| 870 | pjd.inlen = buflen; |
| 871 | pjd.flags = flags; |
| 872 | |
| 873 | int result = pjpeg_decode_buffer(&pjd); |
| 874 | if (error) |
| 875 | *error = result; |
| 876 | |
| 877 | if (result) { |
| 878 | for (int i = 0; i < pjd.ncomponents; i++) |
| 879 | free(pjd.components[i].data); |
| 880 | free(pjd.components); |
| 881 | |
| 882 | return NULL; |
| 883 | } |
| 884 | |
| 885 | pjpeg_t *pj = calloc(1, sizeof(pjpeg_t)); |
| 886 | |
| 887 | pj->width = pjd.width; |
| 888 | pj->height = pjd.height; |
| 889 | pj->ncomponents = pjd.ncomponents; |
| 890 | pj->components = pjd.components; |
| 891 | |
| 892 | return pj; |
| 893 | } |