WebP Image Format

Introduction WebP is a Resource Interchange File Format (RIFF) based image file format () which supports lossless and lossy compression as well as alpha (transparency) and animation. It covers use cases similar to JPEG, PNG and the Graphics Interchange Format (GIF). WebP consists of two compression algorithms used to reduce the size of image pixel data, including alpha (transparency) information. Lossy compression is achieved using VP8 intra-frame encoding . The lossless algorithm stores and restores the pixel values exactly, including the color values for zero alpha pixels. The format uses subresolution images, recursively embedded into the format itself, for storing statistical data about the images, such as the used entropy codes, spatial predictors, color space conversion, and color table. , prefix coding, and a color cache are used for compression of the bulk data.

WebP Container Specification Note this section is based on the documentation in the libwebp source repository.

Introduction WebP is an image format that uses either (i) the VP8 intra-frame encoding to compress image data in a lossy way, or (ii) the WebP lossless encoding. These encoding schemes should make it more efficient than currently used formats. It is optimized for fast image transfer over the network (e.g., for websites). The WebP format has feature parity (color profile, metadata, animation, etc.) with other formats as well. This section describes the structure of a WebP file. The WebP container (i.e., RIFF container for WebP) allows feature support over and above the basic use case of WebP (i.e., a file containing a single image encoded as a VP8 key frame). The WebP container provides additional support for:

Lossless compression. An image can be losslessly compressed, using the WebP Lossless Format.
Metadata. An image may have metadata stored in or formats.
Transparency. An image may have transparency, i.e., an alpha channel.
Color Profile. An image may have an embedded ICC profile.
Animation. An image may have multiple frames with pauses between them, making it an animation.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. Bit numbering in chunk diagrams starts at 0 for the most significant bit ('MSB 0') as described in .

Terminology & Basics A WebP file contains either a still image (i.e., an encoded matrix of pixels) or an animation. Optionally, it can also contain transparency information, color profile and metadata. In case we need to refer only to the matrix of pixels, we will call it the canvas of the image. Below are additional terms used throughout this section:

Reader/Writer: Code that reads WebP files is referred to as a reader, while code that writes them is referred to as a writer.
uint16: A 16-bit, little-endian, unsigned integer.
uint24: A 24-bit, little-endian, unsigned integer.
uint32: A 32-bit, little-endian, unsigned integer.
FourCC: A FourCC (four-character code) is a uint32 created by concatenating four ASCII characters in little-endian order. This means 'aaaa' (0x61616161) and 'AAAA' (0x41414141) are treated as different FourCCs.
1-based: An unsigned integer field storing values offset by -1. e.g., Such a field would store value 25 as 24.
ChunkHeader('ABCD'): This is used to describe the FourCC and Chunk Size header of individual chunks, where 'ABCD' is the FourCC for the chunk. This element's size is 8 bytes.

RIFF File Format The WebP file format is based on the RIFF (Resource Interchange File Format) document format. The basic element of a RIFF file is a chunk. It consists of:

RIFF chunk structure

Chunk FourCC: 32 bits: ASCII four-character code used for chunk identification.
Chunk Size: 32 bits (uint32): The size of the chunk in bytes, not including this field, the chunk identifier or padding.
Chunk Payload: Chunk Size bytes: The data payload. If Chunk Size is odd, a single padding byte -- that MUST be 0 to conform with RIFF -- is added.

Note: RIFF has a convention that all-uppercase chunk FourCCs are standard chunks that apply to any RIFF file format, while FourCCs specific to a file format are all lowercase. WebP does not follow this convention.

WebP File Header

WebP file header chunk

'RIFF': 32 bits: The ASCII characters 'R' 'I' 'F' 'F'.
File Size: 32 bits (uint32): The size of the file in bytes starting at offset 8. The maximum value of this field is 2^32 minus 10 bytes and thus the size of the whole file is at most 4GiB minus 2 bytes.
'WEBP': 32 bits: The ASCII characters 'W' 'E' 'B' 'P'.

A WebP file MUST begin with a RIFF header with the FourCC 'WEBP'. The file size in the header is the total size of the chunks that follow plus 4 bytes for the 'WEBP' FourCC. The file SHOULD NOT contain any data after the data specified by File Size. Readers MAY parse such files, ignoring the trailing data. As the size of any chunk is even, the size given by the RIFF header is also even. The contents of individual chunks will be described in the following sections.

Simple File Format (Lossy) This layout SHOULD be used if the image requires lossy encoding and does not require transparency or other advanced features provided by the extended format. Files with this layout are smaller and supported by older software. Simple WebP (lossy) file format:

Simple WebP (lossy) file format VP8 chunk:

VP8 chunk

VP8 data: Chunk Size bytes: VP8 bitstream data.

Note the fourth character in the 'VP8 ' FourCC is an ASCII space (0x20). The VP8 bitstream format specification is described by . Note that the VP8 frame header contains the VP8 frame width and height. That is assumed to be the width and height of the canvas. The VP8 specification describes how to decode the image into Y'CbCr format. To convert to RGB, Rec. 601 SHOULD be used. Applications MAY use another conversion method, but visual results may differ among decoders.

Simple File Format (Lossless) Note: Older readers may not support files using the lossless format. This layout SHOULD be used if the image requires lossless encoding (with an optional transparency channel) and does not require advanced features provided by the extended format. Simple WebP (lossless) file format:

Simple WebP (lossless) file format VP8L chunk:

VP8L chunk

VP8L data: Chunk Size bytes: VP8L bitstream data.

The specification of the VP8L bitstream can be found in . Note that the VP8L header contains the VP8L image width and height. That is assumed to be the width and height of the canvas.

Extended File Format Note: Older readers may not support files using the extended format. An extended format file consists of:

A 'VP8X' chunk with information about features used in the file.
An optional 'ICCP' chunk with color profile.
An optional 'ANIM' chunk with animation control data.
Image data.
An optional 'EXIF' chunk with Exif metadata.
An optional 'XMP ' chunk with XMP metadata.
An optional list of unknown chunks.

For a still image, the image data consists of a single frame, which is made up of:

An optional alpha subchunk.
A bitstream subchunk.

For an animated image, the image data consists of multiple frames. More details about frames can be found in . All chunks SHOULD be placed in the same order as listed above. If a chunk appears in the wrong place, the file is invalid, but readers MAY parse the file, ignoring the chunks that are out of order. Rationale: Setting the order of chunks should allow quicker file parsing. For example, if an 'ALPH' chunk does not appear in its required position, a decoder can choose to stop searching for it. The rule of ignoring late chunks should make programs that need to do a full search give the same results as the ones stopping early. Extended WebP file header:

Extended WebP file header

Reserved (Rsv): 2 bits: MUST be 0. Readers MUST ignore this field.
ICC profile (I): 1 bit: Set if the file contains an ICC profile.
Alpha (L): 1 bit: Set if any of the frames of the image contain transparency information ("alpha").
Exif metadata (E): 1 bit: Set if the file contains Exif metadata.
XMP metadata (X): 1 bit: Set if the file contains XMP metadata.
Animation (A): 1 bit: Set if this is an animated image. Data in 'ANIM' and 'ANMF' chunks should be used to control the animation.
Reserved (R): 1 bit: MUST be 0. Readers MUST ignore this field.
Reserved: 24 bits: MUST be 0. Readers MUST ignore this field.
Canvas Width Minus One: 24 bits: 1-based width of the canvas in pixels. The actual canvas width is 1 + Canvas Width Minus One
Canvas Height Minus One: 24 bits: 1-based height of the canvas in pixels. The actual canvas height is 1 + Canvas Height Minus One

The product of Canvas Width and Canvas Height MUST be at most 2^32 - 1. Future specifications may add more fields. Unknown fields MUST be ignored.

Chunks

Animation An animation is controlled by ANIM and ANMF chunks. ANIM Chunk: For an animated image, this chunk contains the global parameters of the animation.

ANIM chunk

Background Color: 32 bits (uint32)

The default background color of the canvas in [Blue, Green, Red, Alpha] byte order. This color MAY be used to fill the unused space on the canvas around the frames, as well as the transparent pixels of the first frame. Background color is also used when disposal method is 1. Note:

Background color MAY contain a non-opaque alpha value, even if the Alpha flag in VP8X chunk is unset.
Viewer applications SHOULD treat the background color value as a hint, and are not required to use it.
The canvas is cleared at the start of each loop. The background color MAY be used to achieve this.

Loop Count: 16 bits (uint16)

The number of times to loop the animation. 0 means infinitely.

This chunk MUST appear if the Animation flag in the VP8X chunk is set. If the Animation flag is not set and this chunk is present, it MUST be ignored. ANMF chunk: For animated images, this chunk contains information about a single frame. If the Animation flag is not set, then this chunk SHOULD NOT be present.

ANMF chunk

Frame X: 24 bits (uint24)

The X coordinate of the upper left corner of the frame is Frame X * 2

Frame Y: 24 bits (uint24)

The Y coordinate of the upper left corner of the frame is Frame Y * 2

Frame Width Minus One: 24 bits (uint24)

The 1-based width of the frame. The frame width is 1 + Frame Width Minus One

Frame Height Minus One: 24 bits (uint24)

The 1-based height of the frame. The frame height is 1 + Frame Height Minus One

Frame Duration: 24 bits (uint24)

The time to wait before displaying the next frame, in 1 millisecond units. Note the interpretation of frame duration of 0 (and often <= 10) is implementation defined. Many tools and browsers assign a minimum duration similar to GIF.

Reserved: 6 bits

MUST be 0. Readers MUST ignore this field.

Blending method (B): 1 bit

Indicates how transparent pixels of the current frame are to be blended with corresponding pixels of the previous canvas:

0: Use alpha blending. After disposing of the previous frame, render the current frame on the canvas using alpha-blending. If the current frame does not have an alpha channel, assume alpha value of 255, effectively replacing the rectangle.
1: Do not blend. After disposing of the previous frame, render the current frame on the canvas by overwriting the rectangle covered by the current frame.

Disposal method (D): 1 bit

Indicates how the current frame is to be treated after it has been displayed (before rendering the next frame) on the canvas:

0: Do not dispose. Leave the canvas as is.
1: Dispose to background color. Fill the rectangle on the canvas covered by the current frame with background color specified in the ANIM chunk.

Notes:

The frame disposal only applies to the frame rectangle, that is, the rectangle defined by Frame X, Frame Y, frame width and frame height. It may or may not cover the whole canvas.
Alpha-blending: Given that each of the R, G, B and A channels is 8-bit, and the RGB channels are not premultiplied by alpha, the formula for blending 'dst' onto 'src' is:
Alpha-blending SHOULD be done in linear color space, by taking into account the color profile of the image. If the color profile is not present, sRGB is to be assumed. (Note that sRGB also needs to be linearized due to a gamma of ~2.2).

Frame Data: Chunk Size - 16 bytes

Consists of:

An optional alpha subchunk for the frame.
A bitstream subchunk for the frame.
An optional list of unknown chunks.

Note: The 'ANMF' payload, Frame Data above, consists of individual padded chunks as described by the RIFF file format.

Alpha

ALPH chunk

Reserved (Rsv): 2 bits

MUST be 0. Readers MUST ignore this field.

Pre-processing (P): 2 bits

These informative bits are used to signal the pre-processing that has been performed during compression. The decoder can use this information to e.g. dither the values or smooth the gradients prior to display.

0: No pre-processing.
1: Level reduction.

Filtering method (F): 2 bits

The filtering method used:

0: None.
1: Horizontal filter.
2: Vertical filter.
3: Gradient filter.

For each pixel, filtering is performed using the following calculations. Assume the alpha values surrounding the current X position are labeled as:

We seek to compute the alpha value at position X. First, a prediction is made depending on the filtering method:

Method 0: predictor = 0
Method 1: predictor = A
Method 2: predictor = B
Method 3: predictor = clip(A + B - C)

where clip(v) is equal to:

0 if v < 0
255 if v > 255
v otherwise

The final value is derived by adding the decompressed value X to the predictor and using modulo-256 arithmetic to wrap the [256..511] range into the [0..255] one: There are special cases for the left-most and top-most pixel positions:

The top-left value at location (0, 0) uses 0 as predictor value. Otherwise,
For horizontal or gradient filtering methods, the left-most pixels at location (0, y) are predicted using the location (0, y-1) just above.
For vertical or gradient filtering methods, the top-most pixels at location (x, 0) are predicted using the location (x-1, 0) on the left.

Decoders are not required to use this information in any specified way.

Compression method (C): 2 bits

The compression method used:

0: No compression.
1: Compressed using the WebP lossless format.

Alpha bitstream: Chunk Size - 1 bytes

Encoded alpha bitstream.

This optional chunk contains encoded alpha data for this frame. A frame containing a 'VP8L' chunk SHOULD NOT contain this chunk. Rationale: The transparency information is already part of the 'VP8L' chunk. The alpha channel data is stored as uncompressed raw data (when compression method is '0') or compressed using the lossless format (when the compression method is '1').

Raw data: consists of a byte sequence of length width * height, containing all the 8-bit transparency values in scan order.
Lossless format compression: the byte sequence is a compressed image-stream (as described in ) of implicit dimension width x height. That is, this image-stream does NOT contain any headers describing the image dimension. Rationale: the dimension is already known from other sources, so storing it again would be redundant and error-prone. Once the image-stream is decoded into ARGB color values, following the process described in the lossless format specification, the transparency information must be extracted from the green channel of the ARGB quadruplet. Rationale: the green channel is allowed extra transformation steps in the specification -- unlike the other channels -- that can improve compression.

Bitstream (VP8/VP8L) This chunk contains compressed bitstream data for a single frame. A bitstream chunk may be either (i) a VP8 chunk, using "VP8 " (note the significant fourth-character space) as its tag or (ii) a VP8L chunk, using "VP8L" as its tag. The formats of VP8 and VP8L chunks are as described in and respectively.

Color Profile

ICCP chunk

Color Profile: Chunk Size bytes: ICC profile.

This chunk MUST appear before the image data. There SHOULD be at most one such chunk. If there are more such chunks, readers MAY ignore all except the first one. See the ICC Specification for details. If this chunk is not present, sRGB SHOULD be assumed.

Metadata Metadata can be stored in 'EXIF' or 'XMP ' chunks. There SHOULD be at most one chunk of each type ('EXIF' and 'XMP '). If there are more such chunks, readers MAY ignore all except the first one. The chunks are defined as follows: EXIF chunk:

EXIF chunk

Exif Metadata: Chunk Size bytes: Image metadata in format.

XMP chunk:

XMP chunk

XMP Metadata: Chunk Size bytes: Image metadata in format.

Note the fourth character in the 'XMP ' FourCC is an ASCII space (0x20). Additional guidance about handling metadata can be found in the Metadata Working Group's Guidelines for Handling Metadata.

Unknown Chunks A RIFF chunk (described in .) whose chunk tag is different from any of the chunks described in this section, is considered an unknown chunk. Rationale: Allowing unknown chunks gives a provision for future extension of the format, and also allows storage of any application-specific data. A file MAY contain unknown chunks:

At the end of the file as described in .
At the end of ANMF chunks as described in .

Readers SHOULD ignore these chunks. Writers SHOULD preserve them in their original order (unless they specifically intend to modify these chunks).

Assembling the Canvas From Frames Here we provide an overview of how a reader MUST assemble a canvas in the case of an animated image. The process begins with creating a canvas using the dimensions given in the 'VP8X' chunk, Canvas Width Minus One + 1 pixels wide by Canvas Height Minus One + 1 pixels high. The Loop Count field from the 'ANIM' chunk controls how many times the animation process is repeated. This is Loop Count - 1 for non-zero Loop Count values or infinitely if Loop Count is zero. At the beginning of each loop iteration the canvas is filled using the background color from the 'ANIM' chunk or an application defined color. 'ANMF' chunks contain individual frames given in display order. Before rendering each frame, the previous frame's Disposal method is applied. The rendering of the decoded frame begins at the Cartesian coordinates (2 * Frame X, 2 * Frame Y) using the top-left corner of the canvas as the origin. Frame Width Minus One + 1 pixels wide by Frame Height Minus One + 1 pixels high are rendered onto the canvas using the Blending method. The canvas is displayed for Frame Duration milliseconds. This continues until all frames given by 'ANMF' chunks have been displayed. A new loop iteration is then begun or the canvas is left in its final state if all iterations have been completed. The following pseudocode illustrates the rendering process. The notation VP8X.field means the field in the 'VP8X' chunk with the same description. = frame_right assert VP8X.canvasHeight >= frame_bottom for subchunk in 'Frame Data': if subchunk.tag == "ALPH": assert alpha subchunks not found in 'Frame Data' earlier frame_params.alpha = alpha_data else if subchunk.tag == "VP8 " OR subchunk.tag == "VP8L": assert bitstream subchunks not found in 'Frame Data' earlier frame_params.bitstream = bitstream_data render frame with frame_params.alpha and frame_params.bitstream on canvas with top-left corner at (frame_params.frameX, frame_params.frameY), using blending method frame_params.blendingMethod. canvas contains the decoded image. Show the contents of the canvas for frame_params.frameDuration * 1ms. dispose_method = frame_params.disposeMethod ]]>

Example File Layouts A lossy encoded image with alpha may look as follows:

A losslessly encoded image may look as follows:

A lossless image with ICC profile and XMP metadata may look as follows:

An animated image with Exif metadata may look as follows:

Specification for WebP Lossless Bitstream Note this section is based on the documentation in the libwebp source repository.

Abstract WebP lossless is an image format for lossless compression of ARGB images. The lossless format stores and restores the pixel values exactly, including the color values for pixels whose alpha value is 0. The format uses subresolution images, recursively embedded into the format itself, for storing statistical data about the images, such as the used entropy codes, spatial predictors, color space conversion, and color table. LZ77, prefix coding, and a color cache are used for compression of the bulk data. Decoding speeds faster than PNG have been demonstrated, as well as 25% denser compression than can be achieved using today's PNG format .

Introduction This section describes the compressed data representation of a WebP lossless image. In this section, we extensively use C programming language syntax to describe the bitstream, and assume the existence of a function for reading bits, ReadBits(n). The bytes are read in the natural order of the stream containing them, and bits of each byte are read in least-significant-bit-first order. When multiple bits are read at the same time, the integer is constructed from the original data in the original order. The most significant bits of the returned integer are also the most significant bits of the original data. Thus, the statement is equivalent with the two statements below: We assume that each color component (e.g. alpha, red, blue and green) is represented using an 8-bit byte. We define the corresponding type as uint8. A whole ARGB pixel is represented by a type called uint32, an unsigned integer consisting of 32 bits. In the code showing the behavior of the transformations, alpha value is codified in bits 31..24, red in bits 23..16, green in bits 15..8 and blue in bits 7..0, but implementations of the format are free to use another representation internally. Broadly, a WebP lossless image contains header data, transform information and actual image data. Headers contain width and height of the image. A WebP lossless image can go through four different types of transformation before being entropy encoded. The transform information in the bitstream contains the data required to apply the respective inverse transforms.

Nomenclature

ARGB: A pixel value consisting of alpha, red, green, and blue values.
ARGB image: A two-dimensional array containing ARGB pixels.
color cache: A small hash-addressed array to store recently used colors, to be able to recall them with shorter codes.
color indexing image: A one-dimensional image of colors that can be indexed using a small integer (up to 256 within WebP lossless).
color transform image: A two-dimensional subresolution image containing data about correlations of color components.
distance mapping: Changes LZ77 distances to have the smallest values for pixels in 2D proximity.
entropy image: A two-dimensional subresolution image indicating which entropy coding should be used in a respective square in the image, i.e., each pixel is a meta prefix code.
prefix code: A classic way to do entropy coding where a smaller number of bits are used for more frequent codes.
: Dictionary-based sliding window compression algorithm that either emits symbols or describes them as sequences of past symbols.
meta prefix code: A small integer (up to 16 bits) that indexes an element in the meta prefix table.
predictor image: A two-dimensional subresolution image indicating which spatial predictor is used for a particular square in the image.
prefix coding: A way to entropy code larger integers that codes a few bits of the integer using an entropy code and codifies the remaining bits raw. This allows for the descriptions of the entropy codes to remain relatively small even when the range of symbols is large.
scan-line order: A processing order of pixels, left-to-right, top-to-bottom, starting from the left-hand-top pixel, proceeding to the right. Once a row is completed, continue from the left-hand column of the next row.

RIFF Header The beginning of the header has the RIFF container. This consists of the following 21 bytes:

String "RIFF"
A little-endian 32 bit value of the block length, the whole size of the block controlled by the RIFF header. Normally this equals the payload size (file size minus 8 bytes: 4 bytes for the 'RIFF' identifier and 4 bytes for storing the value itself).
String "WEBP" (RIFF container name).
String "VP8L" (chunk tag for lossless encoded image data).
A little-endian 32-bit value of the number of bytes in the lossless stream.
One byte signature 0x2f.

The first 28 bits of the bitstream specify the width and height of the image. Width and height are decoded as 14-bit integers as follows: The 14-bit dynamics for image size limit the maximum size of a WebP lossless image to 16384x16384 pixels. The alpha_is_used bit is a hint only, and SHOULD NOT impact decoding. It SHOULD be set to 0 when all alpha values are 255 in the picture, and 1 otherwise. The version_number is a 3 bit code that MUST be set to 0. Any other value MUST be treated as an error.

Transformations Transformations are reversible manipulations of the image data that can reduce the remaining symbolic entropy by modeling spatial and color correlations. Transformations can make the final compression more dense. An image can go through four types of transformation. A 1 bit indicates the presence of a transform. Each transform is allowed to be used only once. The transformations are used only for the main level ARGB image: the subresolution images have no transforms, not even the 0 bit indicating the end-of-transforms. Typically, an encoder would use these transforms to reduce the Shannon entropy in the residual image. Also, the transform data can be decided based on entropy minimization. If a transform is present then the next two bits specify the transform type. There are four types of transforms. The transform type is followed by the transform data. Transform data contains the information required to apply the inverse transform and depends on the transform type. Next we describe the transform data for different types.

Predictor Transform The predictor transform can be used to reduce entropy by exploiting the fact that neighboring pixels are often correlated. In the predictor transform, the current pixel value is predicted from the pixels already decoded (in scan-line order) and only the residual value (actual - predicted) is encoded. The prediction mode determines the type of prediction to use. We divide the image into squares and all the pixels in a square use the same prediction mode. The first 3 bits of prediction data define the block width and height in number of bits. The number of block columns, block_xsize, is used in indexing two-dimensionally. The transform data contains the prediction mode for each block of the image. All the block_width * block_height pixels of a block use same prediction mode. The prediction modes are treated as pixels of an image and encoded using the same techniques described in . For a pixel x, y, one can compute the respective filter block address by: > size_bits) * block_xsize + (x >> size_bits); ]]> There are 14 different prediction modes. In each prediction mode, the current pixel value is predicted from one or more neighboring pixels whose values are already known. We choose the neighboring pixels (TL, T, TR, and L) of the current pixel (P) as follows:

where TL means top-left, T top, TR top-right, L left pixel. At the time of predicting a value for P, all pixels O, TL, T, TR and L have already been processed, and pixel P and all pixels X are unknown. Given the above neighboring pixels, the different prediction modes are defined as follows.

Average2 is defined as follows for each ARGB component: The Select predictor is defined as follows: The functions ClampAddSubtractFull and ClampAddSubtractHalf are performed for each ARGB component as follows: 255) ? 255 : a; } int ClampAddSubtractFull(int a, int b, int c) { return Clamp(a + b - c); } int ClampAddSubtractHalf(int a, int b) { return Clamp(a + (a - b) / 2); } ]]> There are special handling rules for some border pixels. If there is a prediction transform, regardless of the mode [0..13] for these pixels, the predicted value for the left-topmost pixel of the image is 0xff000000, L-pixel for all pixels on the top row, and T-pixel for all pixels on the leftmost column. Addressing the TR-pixel for pixels on the rightmost column is exceptional. The pixels on the rightmost column are predicted by using the modes [0..13] just like pixels not on the border, but the leftmost pixel on the same row as the current pixel is instead used as the TR-pixel.

Color Transform The goal of the color transform is to decorrelate the R, G and B values of each pixel. The color transform keeps the green (G) value as it is, transforms red (R) based on green and transforms blue (B) based on green and then based on red. As is the case for the predictor transform, first the image is divided into blocks and the same transform mode is used for all the pixels in a block. For each block there are three types of color transform elements. The actual color transformation is done by defining a color transform delta. The color transform delta depends on the ColorTransformElement, which is the same for all the pixels in a particular block. The delta is subtracted during the color transform. The inverse color transform then is just adding those deltas. The color transform function is defined as follows: green_to_red_, green); tmp_blue -= ColorTransformDelta(trans->green_to_blue_, green); tmp_blue -= ColorTransformDelta(trans->red_to_blue_, red); *new_red = tmp_red & 0xff; *new_blue = tmp_blue & 0xff; } ]]> ColorTransformDelta is computed using a signed 8-bit integer representing a 3.5-fixed-point number, and a signed 8-bit RGB color channel (c) [-128..127] and is defined as follows: > 5; } ]]> A conversion from the 8-bit unsigned representation (uint8) to the 8-bit signed one (int8) is required before calling ColorTransformDelta(). It should be performed using 8-bit two's complement (that is: uint8 range [128..255] is mapped to the [-128..-1] range of its converted int8 value). The multiplication is to be done using more precision (with at least 16-bit dynamics). The sign extension property of the shift operation does not matter here: only the lowest 8 bits are used from the result, and there the sign extension shifting and unsigned shifting are consistent with each other. Now we describe the contents of color transform data so that decoding can apply the inverse color transform and recover the original red and blue values. The first 3 bits of the color transform data contain the width and height of the image block in number of bits, just like the predictor transform: The remaining part of the color transform data contains ColorTransformElement instances corresponding to each block of the image. ColorTransformElement instances are treated as pixels of an image and encoded using the methods described in . During decoding, ColorTransformElement instances of the blocks are decoded and the inverse color transform is applied on the ARGB values of the pixels. As mentioned earlier, that inverse color transform is just adding ColorTransformElement values to the red and blue channels. green_to_red, green); tmp_blue += ColorTransformDelta(trans->green_to_blue, green); tmp_blue += ColorTransformDelta(trans->red_to_blue, tmp_red & 0xff); *new_red = tmp_red & 0xff; *new_blue = tmp_blue & 0xff; } ]]>

Subtract Green Transform The subtract green transform subtracts green values from red and blue values of each pixel. When this transform is present, the decoder needs to add the green value to both red and blue. There is no data associated with this transform. The decoder applies the inverse transform as follows: This transform is redundant as it can be modeled using the color transform, but it is still often useful. Since it can extend the dynamics of the color transform and there is no additional data here, the subtract green transform can be coded using fewer bits than a full-blown color transform.

Color Indexing Transform If there are not many unique pixel values, it may be more efficient to create a color index array and replace the pixel values by the array's indices. The color indexing transform achieves this. (In the context of WebP lossless, we specifically do not call this a palette transform because a similar but more dynamic concept exists in WebP lossless encoding: color cache). The color indexing transform checks for the number of unique ARGB values in the image. If that number is below a threshold (256), it creates an array of those ARGB values, which is then used to replace the pixel values with the corresponding index: the green channel of the pixels are replaced with the index; all alpha values are set to 255; all red and blue values to 0. The transform data contains color table size and the entries in the color table. The decoder reads the color indexing transform data as follows: The color table is stored using the image storage format itself. The color table can be obtained by reading an image, without the RIFF header, image size, and transforms, assuming a height of one pixel and a width of color_table_size. The color table is always subtraction-coded to reduce image entropy. The deltas of palette colors contain typically much less entropy than the colors themselves, leading to significant savings for smaller images. In decoding, every final color in the color table can be obtained by adding the previous color component values by each ARGB component separately, and storing the least significant 8 bits of the result. The inverse transform for the image is simply replacing the pixel values (which are indices to the color table) with the actual color table values. The indexing is done based on the green component of the ARGB color. If the index is equal or larger than color_table_size, the argb color value should be set to 0x00000000 (transparent black). When the color table is small (equal to or less than 16 colors), several pixels are bundled into a single pixel. The pixel bundling packs several (2, 4, or 8) pixels into a single pixel, reducing the image width respectively. Pixel bundling allows for a more efficient joint distribution entropy coding of neighboring pixels, and gives some arithmetic coding-like benefits to the entropy code, but it can only be used when there are 16 or fewer unique values. color_table_size specifies how many pixels are combined: width_bits has a value of 0, 1, 2 or 3. A value of 0 indicates no pixel bundling is to be done for the image. A value of 1 indicates that two pixels are combined, and each pixel has a range of [0..15]. A value of 2 indicates that four pixels are combined, and each pixel has a range of [0..3]. A value of 3 indicates that eight pixels are combined, and each pixel has a range of [0..1], i.e., a binary value. The values are packed into the green component as follows:

width_bits = 1: for every x value where x = 2k + 0, a green value at x is positioned into the 4 least-significant bits of the green value at x / 2, a green value at x + 1 is positioned into the 4 most-significant bits of the green value at x / 2.
width_bits = 2: for every x value where x = 4k + 0, a green value at x is positioned into the 2 least-significant bits of the green value at x / 4, green values at x + 1 to x + 3 are positioned in order to the more significant bits of the green value at x / 4.
width_bits = 3: for every x value where x = 8k + 0, a green value at x is positioned into the least-significant bit of the green value at x / 8, green values at x + 1 to x + 7 are positioned in order to the more significant bits of the green value at x / 8.

Image Data Image data is an array of pixel values in scan-line order.

Roles of Image Data We use image data in five different roles:

ARGB image: Stores the actual pixels of the image.
Entropy image: Stores the meta prefix codes. The red and green components of a pixel define the meta prefix code used in a particular block of the ARGB image.
Predictor image: Stores the metadata for Predictor Transform. The green component of a pixel defines which of the 14 predictors is used within a particular block of the ARGB image.
Color transform image. It is created by ColorTransformElement values (defined in Color Transform) for different blocks of the image. Each ColorTransformElement 'cte' is treated as a pixel whose alpha component is 255, red component is cte.red_to_blue, green component is cte.green_to_blue and blue component is cte.green_to_red.
Color indexing image: An array of size color_table_size (up to 256 ARGB values) storing the metadata for the Color Indexing Transform. This is stored as an image of width color_table_size and height 1.

Encoding of Image Data The encoding of image data is independent of its role. The image is first divided into a set of fixed-size blocks (typically 16x16 blocks). Each of these blocks are modeled using their own entropy codes. Also, several blocks may share the same entropy codes. Rationale: Storing an entropy code incurs a cost. This cost can be minimized if statistically similar blocks share an entropy code, thereby storing that code only once. For example, an encoder can find similar blocks by clustering them using their statistical properties, or by repeatedly joining a pair of randomly selected clusters when it reduces the overall amount of bits needed to encode the image. Each pixel is encoded using one of the three possible methods:

Prefix coded literal: each channel (green, red, blue and alpha) is entropy-coded independently;
LZ77 backward reference: a sequence of pixels are copied from elsewhere in the image; or
Color cache code: using a short multiplicative hash code (color cache index) of a recently seen color.

The following subsections describe each of these in detail.

Prefix Coded Literals The pixel is stored as prefix coded values of green, red, blue and alpha (in that order). See for details.

LZ77 Backward Reference Backward references are tuples of length and distance code:

Length indicates how many pixels in scan-line order are to be copied.
Distance code is a number indicating the position of a previously seen pixel, from which the pixels are to be copied. The exact mapping is described below.

The length and distance values are stored using LZ77 prefix coding. LZ77 prefix coding divides large integer values into two parts: the prefix code and the extra bits: the prefix code is stored using an entropy code, while the extra bits are stored as they are (without an entropy code). Rationale: This approach reduces the storage requirement for the entropy code. Also, large values are usually rare, and so extra bits would be used for very few values in the image. Thus, this approach results in better compression overall. The following table denotes the prefix codes and extra bits used for storing different ranges of values. Note: The maximum backward reference length is limited to 4096. Hence, only the first 24 prefix codes (with the respective extra bits) are meaningful for length values. For distance values, however, all the 40 prefix codes are valid.

The pseudocode to obtain a (length or distance) value from the prefix code is as follows: > 1; int offset = (2 + (prefix_code & 1)) << extra_bits; return offset + ReadBits(extra_bits) + 1; ]]> Distance Mapping: As noted previously, a distance code is a number indicating the position of a previously seen pixel, from which the pixels are to be copied. This subsection defines the mapping between a distance code and the position of a previous pixel. Distance codes larger than 120 denote the pixel-distance in scan-line order, offset by 120. The smallest distance codes [1..120] are special, and are reserved for a close neighborhood of the current pixel. This neighborhood consists of 120 pixels:

Pixels that are 1 to 7 rows above the current pixel, and are up to 8 columns to the left or up to 7 columns to the right of the current pixel. [Total such pixels = 7 * (8 + 1 + 7) = 112].
Pixels that are in same row as the current pixel, and are up to 8 columns to the left of the current pixel. [8 such pixels].

The mapping between distance code i and the neighboring pixel offset (xi, yi) is as follows:

For example, the distance code 1 indicates an offset of (0, 1) for the neighboring pixel, that is, the pixel above the current pixel (0-pixel difference in the X-direction and 1 pixel difference in the Y-direction). Similarly, the distance code 3 indicates the left-top pixel. The decoder can convert a distance code i to a scan-line order distance dist as follows: where distance_map is the mapping noted above and xsize is the width of the image in pixels.

Color Cache Coding Color cache stores a set of colors that have been recently used in the image. Rationale: This way, the recently used colors can sometimes be referred to more efficiently than emitting them using the other two methods (described in and ). Color cache codes are stored as follows. First, there is a 1-bit value that indicates if the color cache is used. If this bit is 0, no color cache codes exist, and they are not transmitted in the prefix code that decodes the green symbols and the length prefix codes. However, if this bit is 1, the color cache size is read next: color_cache_code_bits defines the size of the color_cache by (1 << color_cache_code_bits). The range of allowed values for color_cache_code_bits is [1..11]. Compliant decoders MUST indicate a corrupted bitstream for other values. A color cache is an array of size color_cache_size. Each entry stores one ARGB color. Colors are looked up by indexing them by (0x1e35a7bd * color) >> (32 - color_cache_code_bits). Only one lookup is done in a color cache; there is no conflict resolution. In the beginning of decoding or encoding of an image, all entries in all color cache values are set to zero. The color cache code is converted to this color at decoding time. The state of the color cache is maintained by inserting every pixel, be it produced by backward referencing or as literals, into the cache in the order they appear in the stream.

Entropy Code

Overview Most of the data is coded using a canonical prefix code. Hence, the codes are transmitted by sending the prefix code lengths, as opposed to the actual prefix codes. In particular, the format uses spatially-variant prefix coding. In other words, different blocks of the image can potentially use different entropy codes. Rationale: Different areas of the image may have different characteristics. So, allowing them to use different entropy codes provides more flexibility and potentially better compression.

Details The encoded image data consists of several parts:

Decoding and building the prefix codes
Meta prefix codes
Entropy-coded image data

Decoding and Building the Prefix Codes There are several steps in decoding the prefix codes. Decoding the Code Lengths: This section describes how to read the prefix code lengths from the bitstream. The prefix code lengths can be coded in two ways. The method used is specified by a 1-bit value.

If this bit is 1, it is a simple code length code, and
If this bit is 0, it is a normal code length code.

In both cases, there can be unused code lengths that are still part of the stream. This may be inefficient, but it is allowed by the format. (i) Simple Code Length Code: This variant is used in the special case when only 1 or 2 prefix symbols are in the range [0..255] with code length 1. All other prefix code lengths are implicitly zeros. The first bit indicates the number of symbols: Following are the symbol values. This first symbol is coded using 1 or 8 bits depending on the value of is_first_8bits. The range is [0..1] or [0..255], respectively. The second symbol, if present, is always assumed to be in the range [0..255] and coded using 8 bits. Note: Another special case is when all prefix code lengths are zeros (an empty prefix code). For example, a prefix code for distance can be empty if there are no backward references. Similarly, prefix codes for alpha, red, and blue can be empty if all pixels within the same meta prefix code are produced using the color cache. However, this case doesn't need special handling, as empty prefix codes can be coded as those containing a single symbol 0. (ii) Normal Code Length Code: The code lengths of the prefix code fit in 8 bits and are read as follows. First, num_code_lengths specifies the number of code lengths. If num_code_lengths is > 19, the bitstream is invalid. The code lengths are themselves encoded using prefix codes: lower level code lengths, code_length_code_lengths, first have to be read. The rest of those code_length_code_lengths (according to the order in kCodeLengthCodeOrder) are zeros. Next, if ReadBits(1) == 0, the maximum number of different read symbols is num_code_lengths. Otherwise, it is defined as: A prefix table is then built from code_length_code_lengths and used to read up to max_symbol code lengths.

Code [0..15] indicates literal code lengths.
- Value 0 means no symbols have been coded.
- Values [1..15] indicate the bit length of the respective code.
Code 16 repeats the previous non-zero value [3..6] times, i.e., 3 + ReadBits(2) times. If code 16 is used before a non-zero value has been emitted, a value of 8 is repeated.
Code 17 emits a streak of zeros [3..10], i.e., 3 + ReadBits(3) times.
Code 18 emits a streak of zeros of length [11..138], i.e., 11 + ReadBits(7) times.

Once code lengths are read, a prefix code for each symbol type (A, R, G, B, distance) is formed using their respective alphabet sizes:

G channel: 256 + 24 + color_cache_size
other literals (A,R,B): 256
distance code: 40

Decoding of Meta Prefix Codes As noted earlier, the format allows the use of different prefix codes for different blocks of the image. Meta prefix codes are indexes identifying which prefix codes to use in different parts of the image. Meta prefix codes may be used only when the image is being used in the role of an ARGB image. There are two possibilities for the meta prefix codes, indicated by a 1-bit value:

If this bit is zero, there is only one meta prefix code used everywhere in the image. No more data is stored.
If this bit is one, the image uses multiple meta prefix codes. These meta prefix codes are stored as an entropy image (described below).

Entropy image: The entropy image defines which prefix codes are used in different parts of the image, as described below. The first 3-bits contain the prefix_bits value. The dimensions of the entropy image are derived from prefix_bits. where DIV_ROUND_UP is as defined in . The next bits contain an entropy image of width prefix_xsize and height prefix_ysize. Interpretation of Meta Prefix Codes: For any given pixel (x, y), there is a set of five prefix codes associated with it. These codes are (in bitstream order):

Prefix code #1: used for green channel, backward-reference length and color cache.
Prefix code #2, #3 and #4: used for red, blue and alpha channels respectively.
Prefix code #5: used for backward-reference distance.

From here on, we refer to this set as a prefix code group. The number of prefix code groups in the ARGB image can be obtained by finding the largest meta prefix code from the entropy image: where max(entropy image) indicates the largest prefix code stored in the entropy image. As each prefix code group contains five prefix codes, the total number of prefix codes is: Given a pixel (x, y) in the ARGB image, we can obtain the corresponding prefix codes to be used as follows: > prefix_bits) * prefix_xsize + (x >> prefix_bits); int meta_prefix_code = (entropy_image[pos] >> 8) & 0xffff; PrefixCodeGroup prefix_group = prefix_code_groups[meta_prefix_code]; ]]> where, we have assumed the existence of PrefixCodeGroup structure, which represents a set of five prefix codes. Also, prefix_code_groups is an array of PrefixCodeGroup (of size num_prefix_groups). The decoder then uses prefix code group prefix_group to decode the pixel (x, y) as explained in .

Decoding Entropy-coded Image Data For the current position (x, y) in the image, the decoder first identifies the corresponding prefix code group (as explained in the last section). Given the prefix code group, the pixel is read and decoded as follows: Read next the symbol S from the bitstream using prefix code #1. Note that S is any integer in the range 0 to (256 + 24 + color_cache_size- 1). The interpretation of S depends on its value:

if S < 256
1. Use S as the green component.
2. Read red from the bitstream using prefix code #2.
3. Read blue from the bitstream using prefix code #3.
4. Read alpha from the bitstream using prefix code #4.
if S >= 256 && S < 256 + 24
1. Use S - 256 as a length prefix code.
2. Read extra bits for length from the bitstream.
3. Determine backward-reference length L from length prefix code and the extra bits read.
4. Read distance prefix code from the bitstream using prefix code #5.
5. Read extra bits for distance from the bitstream.
6. Determine backward-reference distance D from distance prefix code and the extra bits read.
7. Copy the L pixels (in scan-line order) from the sequence of pixels prior to them by D pixels.
if S >= 256 + 24
1. Use S - (256 + 24) as the index into the color cache.
2. Get ARGB color from the color cache at that index.

Overall Structure of the Format Below is a view into the format in Augmented Backus-Naur Form . It does not cover all details. End-of-image (EOI) is only implicitly coded into the number of pixels (xsize * ysize).

Basic Structure

Structure of Transforms

Structure of the Image Data A possible example sequence:

Security Considerations Implementations of this format face security risks such as integer overflows, out-of-bounds reads and writes to both heap and stack, uninitialized data usage, null pointer dereferences, resource (disk, memory) exhaustion and extended resource usage (long running time) as part of the demuxing and decoding process. In particular, implementations reading this format are likely to take input from unknown and possibly unsafe sources -- both clients (e.g., web browsers, email clients) and servers (e.g., applications which accept uploaded images). These may result in arbitrary code execution, information leakage (memory layout and contents) or crashes and thereby allow a device to be compromised or cause a denial of service to an application using the format . The format does not employ "active content", but does allow metadata (, ) and custom chunks to be embedded in a file. Applications that interpret these chunks may be subject to security considerations for those formats.

Interoperability Considerations The format is defined using little-endian byte ordering (see ), but demuxing and decoding are possible on platforms using a different ordering with the appropriate conversion. The container is RIFF-based and allows extension via user defined chunks, but nothing beyond the chunks defined by the container format () are required for decoding of the image. These have been finalized, but were extended in the format's early stages, so some older readers may not support lossless or animated image decoding.

IANA Considerations IANA has registered the following media type .

The 'image/webp' Media Type This section contains the media type registration details as per .

Registration Details RFC Editor Note: Replace RFCXXXX with the published RFC number and add an xref to the entry in the Normative References section. Type name: image Subtype name: webp Required parameters: N/A Optional parameters: N/A Encoding considerations: Binary. The Base64 encoding should be used on transports that cannot accommodate binary data directly. Security considerations: See . Interoperability considerations: See . Published specification: RFCXXXX Applications that use this media type: Applications that are used to display and process images, especially when smaller image file sizes are important. Fragment identifier considerations: N/A Additional information:

Deprecated alias names for this type: N/A
Magic number(s): The first 4 bytes are 0x52, 0x49, 0x46, 0x46 ('RIFF'), followed by 4 bytes for the RIFF chunk size. The next 7 bytes are 0x57, 0x45, 0x42, 0x50, 0x56, 0x50, 0x38 ('WEBPVP8').
File extension(s): webp
Apple Uniform Type Identifier: org.webmproject.webp conforms to public.image
Object Identifiers: N/A

Person & email address to contact for further information:

Name: James Zern
Email: jzern@google.com

Intended usage: COMMON Restrictions on usage: N/A Author:

Name: James Zern
Email: jzern@google.com

Change controller:

Name: James Zern
Email: jzern@google.com
Name: Pascal Massimino
Email: pascal.massimino@gmail.com
Name: WebM Project
Email: webmaster@webmproject.org

Provisional registration? (standards tree only): N/A