4 \__\__/\____/\_____/__/ ____ ___
7 \____/____/\_____/_____/____/v0.3.0
12 WebP codec: library to encode and decode images in WebP format. This package
13 contains the library that can be used in other programs to add WebP support,
14 as well as the command line tools 'cwebp' and 'dwebp'.
16 See http://developers.google.com/speed/webp
18 Latest sources are available from http://www.webmproject.org/code/
20 It is released under the same license as the WebM project.
21 See http://www.webmproject.org/license/software/ or the
22 file "COPYING" file for details. An additional intellectual
23 property rights grant can be found in the file PATENTS.
33 nmake /f Makefile.vc CFG=release-static RTLIBCFG=static OBJDIR=output
35 the directory output\release-static\(x64|x86)\bin will contain the tools
36 cwebp.exe and dwebp.exe. The directory output\release-static\(x64|x86)\lib will
37 contain the libwebp static library.
38 The target architecture (x86/x64) is detected by Makefile.vc from the Visual
39 Studio compiler (cl.exe) available in the system path.
41 Unix build using makefile.unix:
42 -------------------------------
44 On platforms with GNU tools installed (gcc and make), running
48 will build the binaries examples/cwebp and examples/dwebp, along
49 with the static library src/libwebp.a. No system-wide installation
50 is supplied, as this is a simple alternative to the full installation
51 system based on the autoconf tools (see below).
52 Please refer to makefile.unix for additional details and customizations.
56 When building from git sources, you will need to run autogen.sh to generate the
63 should be all you need to have the following files
65 /usr/local/include/webp/decode.h
66 /usr/local/include/webp/encode.h
67 /usr/local/include/webp/types.h
68 /usr/local/lib/libwebp.*
74 Note: A decode-only library, libwebpdecoder, is available using the
75 '--enable-libwebpdecoder' flag. The encode library is built separately and can
76 be installed independently using a minor modification in the corresponding
77 Makefile.am configure files (see comments there). See './configure --help' for
83 To generate language bindings from swig/libwebp.i at least swig-1.3
84 (http://www.swig.org) is required.
86 Currently the following functions are mapped:
102 WebPEncodeLosslessRGBA
103 WebPEncodeLosslessBGRA
104 WebPEncodeLosslessRGB
105 WebPEncodeLosslessBGR
107 See swig/README for more detailed build instructions.
111 To build the swig-generated JNI wrapper code at least JDK-1.5 (or equivalent)
112 is necessary for enum support. The output is intended to be a shared object /
113 DLL that can be loaded via System.loadLibrary("webp_jni").
117 To build the swig-generated Python extension code at least Python 2.6 is
118 required. Python < 2.6 may build with some minor changes to libwebp.i or the
119 generated code, but is untested.
124 The examples/ directory contains tools for encoding (cwebp) and
125 decoding (dwebp) images.
127 The easiest use should look like:
128 cwebp input.png -q 80 -o output.webp
129 which will convert the input file to a WebP file using a quality factor of 80
130 on a 0->100 scale (0 being the lowest quality, 100 being the best. Default
132 You might want to try the -lossless flag too, which will compress the source
133 (in RGBA format) without any loss. The -q quality parameter will in this case
134 control the amount of processing time spent trying to make the output file as
137 A longer list of options is available using the -longhelp command line flag:
141 cwebp [-preset <...>] [options] in_file [-o out_file]
143 If input size (-s) for an image is not specified, it is assumed to be a PNG,
146 -h / -help ............ short help
147 -H / -longhelp ........ long help
148 -q <float> ............. quality factor (0:small..100:big)
149 -alpha_q <int> ......... Transparency-compression quality (0..100).
150 -preset <string> ....... Preset setting, one of:
151 default, photo, picture,
153 -preset must come first, as it overwrites other parameters.
154 -m <int> ............... compression method (0=fast, 6=slowest)
155 -segments <int> ........ number of segments to use (1..4)
156 -size <int> ............ Target size (in bytes)
157 -psnr <float> .......... Target PSNR (in dB. typically: 42)
159 -s <int> <int> ......... Input size (width x height) for YUV
160 -sns <int> ............. Spatial Noise Shaping (0:off, 100:max)
161 -f <int> ............... filter strength (0=off..100)
162 -sharpness <int> ....... filter sharpness (0:most .. 7:least sharp)
163 -strong ................ use strong filter instead of simple (default).
164 -nostrong .............. use simple filter instead of strong.
165 -partition_limit <int> . limit quality to fit the 512k limit on
166 the first partition (0=no degradation ... 100=full)
167 -pass <int> ............ analysis pass number (1..10)
168 -crop <x> <y> <w> <h> .. crop picture with the given rectangle
169 -resize <w> <h> ........ resize picture (after any cropping)
170 -mt .................... use multi-threading if available
171 -low_memory ............ reduce memory usage (slower encoding)
172 -map <int> ............. print map of extra info.
173 -print_psnr ............ prints averaged PSNR distortion.
174 -print_ssim ............ prints averaged SSIM distortion.
175 -print_lsim ............ prints local-similarity distortion.
176 -d <file.pgm> .......... dump the compressed output (PGM file).
177 -alpha_method <int> .... Transparency-compression method (0..1)
178 -alpha_filter <string> . predictive filtering for alpha plane.
179 One of: none, fast (default) or best.
180 -alpha_cleanup ......... Clean RGB values in transparent area.
181 -blend_alpha <hex> ..... Blend colors against background color
182 expressed as RGB values written in
183 hexadecimal, e.g. 0xc0e0d0 for red=0xc0
184 green=0xe0 and blue=0xd0.
185 -noalpha ............... discard any transparency information.
186 -lossless .............. Encode image losslessly.
187 -hint <string> ......... Specify image characteristics hint.
188 One of: photo, picture or graph
190 -metadata <string> ..... comma separated list of metadata to
191 copy from the input to the output if present.
192 Valid values: all, none (default), exif, icc, xmp
194 -short ................. condense printed message
195 -quiet ................. don't print anything.
196 -version ............... print version number and exit.
197 -noasm ................. disable all assembly optimizations.
198 -v ..................... verbose, e.g. print encoding/decoding times
199 -progress .............. report encoding progress
201 Experimental Options:
202 -jpeg_like ............. Roughly match expected JPEG size.
203 -af .................... auto-adjust filter strength.
204 -pre <int> ............. pre-processing filter
207 The main options you might want to try in order to further tune the
215 * 'preset' will set up a default encoding configuration targeting a
216 particular type of input. It should appear first in the list of options,
217 so that subsequent options can take effect on top of this preset.
218 Default value is 'default'.
219 * 'sns' will progressively turn on (when going from 0 to 100) some additional
220 visual optimizations (like: segmentation map re-enforcement). This option
221 will balance the bit allocation differently. It tries to take bits from the
222 "easy" parts of the picture and use them in the "difficult" ones instead.
223 Usually, raising the sns value (at fixed -q value) leads to larger files,
224 but with better quality.
225 Typical value is around '75'.
226 * 'f' option directly links to the filtering strength used by the codec's
227 in-loop processing. The higher the value, the smoother the
228 highly-compressed area will look. This is particularly useful when aiming
229 at very small files. Typical values are around 20-30. Note that using the
230 option -strong/-nostrong will change the type of filtering. Use "-f 0" to
232 * 'm' controls the trade-off between encoding speed and quality. Default is 4.
233 You can try -m 5 or -m 6 to explore more (time-consuming) encoding
234 possibilities. A lower value will result in faster encoding at the expense
240 There is a decoding sample in examples/dwebp.c which will take
241 a .webp file and decode it to a PNG image file (amongst other formats).
242 This is simply to demonstrate the use of the API. You can verify the
243 file test.webp decodes to exactly the same as test_ref.ppm by using:
246 ./dwebp test.webp -ppm -o test.ppm
247 diff test.ppm test_ref.ppm
249 The full list of options is available using -h:
252 Usage: dwebp in_file [options] [-o out_file]
254 Decodes the WebP image file to PNG format [Default]
255 Use following options to convert into alternate image formats:
256 -pam ......... save the raw RGBA samples as a color PAM
257 -ppm ......... save the raw RGB samples as a color PPM
258 -bmp ......... save as uncompressed BMP format
259 -tiff ........ save as uncompressed TIFF format
260 -pgm ......... save the raw YUV samples as a grayscale PGM
261 file with IMC4 layout
262 -yuv ......... save the raw YUV samples in flat layout
265 -version .... print version number and exit.
266 -nofancy ..... don't use the fancy YUV420 upscaler.
267 -nofilter .... disable in-loop filtering.
268 -mt .......... use multi-threading
269 -crop <x> <y> <w> <h> ... crop output with the given rectangle
270 -scale <w> <h> .......... scale the output (*after* any cropping)
271 -alpha ....... only save the alpha plane.
272 -h ....... this help message.
273 -v ....... verbose (e.g. print encoding/decoding times)
274 -noasm ....... disable all assembly optimizations.
279 There's a little self-serve visualization tool called 'vwebp' under the
280 examples/ directory. It uses OpenGL to open a simple drawing window and show
281 a decoded WebP file. It's not yet integrated in the automake build system, but
282 you can try to manually compile it using the recommendations below.
284 Usage: vwebp in_file [options]
286 Decodes the WebP image file and visualize it using OpenGL
288 -version .... print version number and exit.
289 -noicc ....... don't use the icc profile if present.
290 -nofancy ..... don't use the fancy YUV420 upscaler.
291 -nofilter .... disable in-loop filtering.
292 -mt .......... use multi-threading.
293 -info ........ print info.
294 -h ....... this help message.
297 'c' ................ toggle use of color profile.
298 'i' ................ overlay file information.
299 'q' / 'Q' / ESC .... quit.
305 1) OpenGL & OpenGL Utility Toolkit (GLUT)
307 $ sudo apt-get install freeglut3-dev mesa-common-dev
309 - These libraries should be available in the OpenGL / GLUT frameworks.
311 http://freeglut.sourceforge.net/index.php#download
313 2) (Optional) qcms (Quick Color Management System)
314 i. Download qcms from Mozilla / Chromium:
315 http://hg.mozilla.org/mozilla-central/file/0e7639e3bdfb/gfx/qcms
316 http://src.chromium.org/viewvc/chrome/trunk/src/third_party/qcms
317 ii. Build and archive the source files as libqcms.a / qcms.lib
318 iii. Update makefile.unix / Makefile.vc
319 a) Define WEBP_HAVE_QCMS
320 b) Update include / library paths to reference the qcms directory.
322 Build using makefile.unix / Makefile.vc:
323 $ make -f makefile.unix examples/vwebp
324 > nmake /f Makefile.vc CFG=release-static \
325 ../obj/x64/release-static/bin/vwebp.exe
330 The main encoding functions are available in the header src/webp/encode.h
331 The ready-to-use ones are:
332 size_t WebPEncodeRGB(const uint8_t* rgb, int width, int height, int stride,
333 float quality_factor, uint8_t** output);
334 size_t WebPEncodeBGR(const uint8_t* bgr, int width, int height, int stride,
335 float quality_factor, uint8_t** output);
336 size_t WebPEncodeRGBA(const uint8_t* rgba, int width, int height, int stride,
337 float quality_factor, uint8_t** output);
338 size_t WebPEncodeBGRA(const uint8_t* bgra, int width, int height, int stride,
339 float quality_factor, uint8_t** output);
341 They will convert raw RGB samples to a WebP data. The only control supplied
342 is the quality factor.
344 There are some variants for using the lossless format:
346 size_t WebPEncodeLosslessRGB(const uint8_t* rgb, int width, int height,
347 int stride, uint8_t** output);
348 size_t WebPEncodeLosslessBGR(const uint8_t* bgr, int width, int height,
349 int stride, uint8_t** output);
350 size_t WebPEncodeLosslessRGBA(const uint8_t* rgba, int width, int height,
351 int stride, uint8_t** output);
352 size_t WebPEncodeLosslessBGRA(const uint8_t* bgra, int width, int height,
353 int stride, uint8_t** output);
355 Of course in this case, no quality factor is needed since the compression
356 occurs without loss of the input values, at the expense of larger output sizes.
358 Advanced encoding API:
359 ----------------------
361 A more advanced API is based on the WebPConfig and WebPPicture structures.
363 WebPConfig contains the encoding settings and is not tied to a particular
365 WebPPicture contains input data, on which some WebPConfig will be used for
367 The encoding flow looks like:
369 -------------------------------------- BEGIN PSEUDO EXAMPLE
371 #include <webp/encode.h>
373 // Setup a config, starting form a preset and tuning some additional
376 if (!WebPConfigPreset(&config, WEBP_PRESET_PHOTO, quality_factor))
377 return 0; // version error
379 // ... additional tuning
380 config.sns_strength = 90;
381 config.filter_sharpness = 6;
382 config_error = WebPValidateConfig(&config); // not mandatory, but useful
384 // Setup the input data
386 if (!WebPPictureInit(&pic)) {
387 return 0; // version error
391 // allocated picture of dimension width x height
392 if (!WebPPictureAllocate(&pic)) {
393 return 0; // memory error
395 // at this point, 'pic' has been initialized as a container,
396 // and can receive the Y/U/V samples.
397 // Alternatively, one could use ready-made import functions like
398 // WebPPictureImportRGB(), which will take care of memory allocation.
399 // In any case, past this point, one will have to call
400 // WebPPictureFree(&pic) to reclaim memory.
402 // Set up a byte-output write method. WebPMemoryWriter, for instance.
403 WebPMemoryWriter wrt;
404 pic.writer = MyFileWriter;
405 pic.custom_ptr = my_opaque_structure_to_make_MyFileWriter_work;
406 // initialize 'wrt' here...
409 int ok = WebPEncode(&config, &pic); // ok = 0 => error occurred!
410 WebPPictureFree(&pic); // must be called independently of the 'ok' result.
412 // output data should have been handled by the writer at that point.
414 -------------------------------------- END PSEUDO EXAMPLE
419 This is mainly just one function to call:
421 #include "webp/decode.h"
422 uint8_t* WebPDecodeRGB(const uint8_t* data, size_t data_size,
423 int* width, int* height);
425 Please have a look at the file src/webp/decode.h for the details.
426 There are variants for decoding in BGR/RGBA/ARGB/BGRA order, along with
427 decoding to raw Y'CbCr samples. One can also decode the image directly into a
428 pre-allocated buffer.
430 To detect a WebP file and gather the picture's dimensions, the function:
431 int WebPGetInfo(const uint8_t* data, size_t data_size,
432 int* width, int* height);
433 is supplied. No decoding is involved when using it.
435 Incremental decoding API:
436 =========================
438 In the case when data is being progressively transmitted, pictures can still
439 be incrementally decoded using a slightly more complicated API. Decoder state
440 is stored into an instance of the WebPIDecoder object. This object can be
441 created with the purpose of decoding either RGB or Y'CbCr samples.
444 WebPDecBuffer buffer;
445 WebPInitDecBuffer(&buffer);
446 buffer.colorspace = MODE_BGR;
448 WebPIDecoder* idec = WebPINewDecoder(&buffer);
450 As data is made progressively available, this incremental-decoder object
451 can be used to decode the picture further. There are two (mutually exclusive)
452 ways to pass freshly arrived data:
454 either by appending the fresh bytes:
456 WebPIAppend(idec, fresh_data, size_of_fresh_data);
458 or by just mentioning the new size of the transmitted data:
460 WebPIUpdate(idec, buffer, size_of_transmitted_buffer);
462 Note that 'buffer' can be modified between each call to WebPIUpdate, in
463 particular when the buffer is resized to accommodate larger data.
465 These functions will return the decoding status: either VP8_STATUS_SUSPENDED if
466 decoding is not finished yet or VP8_STATUS_OK when decoding is done. Any other
467 status is an error condition.
469 The 'idec' object must always be released (even upon an error condition) by
470 calling: WebPDelete(idec).
472 To retrieve partially decoded picture samples, one must use the corresponding
473 method: WebPIDecGetRGB or WebPIDecGetYUVA.
474 It will return the last displayable pixel row.
476 Lastly, note that decoding can also be performed into a pre-allocated pixel
477 buffer. This buffer must be passed when creating a WebPIDecoder, calling
478 WebPINewRGB() or WebPINewYUVA().
480 Please have a look at the src/webp/decode.h header for further details.
482 Advanced Decoding API:
483 ======================
485 WebP decoding supports an advanced API which provides on-the-fly cropping and
486 rescaling, something of great usefulness on memory-constrained environments like
487 mobile phones. Basically, the memory usage will scale with the output's size,
488 not the input's, when one only needs a quick preview or a zoomed in portion of
489 an otherwise too-large picture. Some CPU can be saved too, incidentally.
491 -------------------------------------- BEGIN PSEUDO EXAMPLE
492 // A) Init a configuration object
493 WebPDecoderConfig config;
494 CHECK(WebPInitDecoderConfig(&config));
496 // B) optional: retrieve the bitstream's features.
497 CHECK(WebPGetFeatures(data, data_size, &config.input) == VP8_STATUS_OK);
499 // C) Adjust 'config' options, if needed
500 config.options.no_fancy_upsampling = 1;
501 config.options.use_scaling = 1;
502 config.options.scaled_width = scaledWidth();
503 config.options.scaled_height = scaledHeight();
506 // D) Specify 'config' output options for specifying output colorspace.
507 // Optionally the external image decode buffer can also be specified.
508 config.output.colorspace = MODE_BGRA;
509 // Optionally, the config.output can be pointed to an external buffer as
510 // well for decoding the image. This externally supplied memory buffer
511 // should be big enough to store the decoded picture.
512 config.output.u.RGBA.rgba = (uint8_t*) memory_buffer;
513 config.output.u.RGBA.stride = scanline_stride;
514 config.output.u.RGBA.size = total_size_of_the_memory_buffer;
515 config.output.is_external_memory = 1;
517 // E) Decode the WebP image. There are two variants w.r.t decoding image.
518 // The first one (E.1) decodes the full image and the second one (E.2) is
519 // used to incrementally decode the image using small input buffers.
520 // Any one of these steps can be used to decode the WebP image.
522 // E.1) Decode full image.
523 CHECK(WebPDecode(data, data_size, &config) == VP8_STATUS_OK);
525 // E.2) Decode image incrementally.
526 WebPIDecoder* const idec = WebPIDecode(NULL, NULL, &config);
528 while (bytes_remaining > 0) {
529 VP8StatusCode status = WebPIAppend(idec, input, bytes_read);
530 if (status == VP8_STATUS_OK || status == VP8_STATUS_SUSPENDED) {
531 bytes_remaining -= bytes_read;
538 // F) Decoded image is now in config.output (and config.output.u.RGBA).
539 // It can be saved, displayed or otherwise processed.
541 // G) Reclaim memory allocated in config's object. It's safe to call
542 // this function even if the memory is external and wasn't allocated
544 WebPFreeDecBuffer(&config.output);
546 -------------------------------------- END PSEUDO EXAMPLE
551 Please report all bugs to our issue tracker:
552 http://code.google.com/p/webp/issues
553 Patches welcome! See this page to get started:
554 http://www.webmproject.org/code/contribute/submitting-patches/
559 Email: webp-discuss@webmproject.org
560 Web: http://groups.google.com/a/webmproject.org/group/webp-discuss