simplejpeg 1.3.9

A simple package for fast JPEG encoding and decoding.

This project is in no way affiliated with the libjpeg-turbo project.

simplejpeg

simplejpeg is a simple package based on recent versions of libturbojpeg for fast JPEG encoding and decoding.

Why another library?

Pillow and OpenCV are excellent options for handling JPEG images and a variety of other formats.

If all you want is to read or write a couple of images and don’t worry about the details, this package is not for you.

Keep reading if you care about speed and want more control over how your JPEGs are handled.

These are the reasons why I started making this:

1. Pillow is very slow compared to OpenCV.

2. Pillow only accepts streams as input. Images in memory have to be wrapped in BytesIO or similar. This adds to the slowness.

3. OpenCV is gigantic, only accepts Numpy arrays as input, and returns images as BGR instead of RGB.

4. Recent versions of libturbojpeg offer impressive speed gains on modern processors. Linux distributions and libraries tend to ship very old versions.

This library is especially for you if you need:

1. Speed.

2. Read and write directly from/to memory.

3. Advanced features of the underlying library.

Installation

• On Linux (x86/x64), Windows (x86/x64), or MacOS (10.9+, x64) you can simply pip install simplejpeg. Update pip if it wants to build from source anyway.

• On other platforms you can try to install from source. Make sure your system is setup to build CPython extensions and install cmake >= 2.8.12. Then run pip install simplejpeg to install from source.

• You can also run python setup.py bdist_wheel etc. as usual.

Usage

This library provides four functions:

decode_jpeg_header, decode_jpeg, encode_jpeg, is_jpeg.

Uncompressed image data is stored as numpy arrays. Decoding functions can accept any Python object that supports the buffer protocol, like array, bytes, bytearray, memoryview, etc.

decode_jpeg_header

decode_jpeg_header(
    data: Any,
    min_height: SupportsInt=0,
    min_width: SupportsInt=0,
    min_factor: SupportsFloat=1,
) -> (SupportsInt, SupportsInt, Text, Text)

Decode only the header of a JPEG image given as JPEG (JFIF) data from memory. Accepts any input that supports the buffer protocol. This is very fast on the order of 100000+ images per second. Returns height and width in pixels of the image when decoded, and colorspace and subsampling as string.

• data: JPEG data in memory; must support buffer interface (e.g., bytes, memoryview)

• min_height: minimum height in pixels of the decoded image; values <= 0 are ignored

• min_width: minimum width in pixels of the decoded image; values <= 0 are ignored

• min_factor: minimum downsampling factor when decoding to smaller size; factors smaller than 2 may take longer to decode

• returns: (height: int, width: int, colorspace: str, color subsampling: str)

decode_jpeg

def decode_jpeg(
    data: SupportsBuffer,
    colorspace: Text='RGB',
    fastdct: Any=False,
    fastupsample: Any=False,
    min_height: SupportsInt=0,
    min_width: SupportsInt=0,
    min_factor: SupportsFloat=1,
    buffer: SupportsBuffer=None,
) -> np.ndarray

Decode a JPEG image given as JPEG (JFIF) data from memory. Accepts any input that supports the buffer protocol. Returns the image as numpy array in the requested colorspace.

• data: JPEG data in memory; must support buffer interface (e.g., bytes, memoryview)

• colorspace: target colorspace, any of the following: ‘RGB’, ‘BGR’, ‘RGBX’, ‘BGRX’, ‘XBGR’, ‘XRGB’, ‘GRAY’, ‘RGBA’, ‘BGRA’, ‘ABGR’, ‘ARGB’; ‘CMYK’ may only be used for images already in CMYK space

• fastdct: if True, use fastest DCT method; speeds up decoding by 4-5% for a minor loss in quality

• fastupsample: if True, use fastest color upsampling method; speeds up decoding by 4-5% for a minor loss in quality

• min_height: minimum height in pixels of the decoded image; values <= 0 are ignored

• param min_width: minimum width in pixels of the decoded image; values <= 0 are ignored

• param min_factor: minimum downsampling factor when decoding to smaller size; factors smaller than 2 may take longer to decode

• buffer: use given object as output buffer; must support the buffer protocol and be writable, e.g., numpy ndarray or bytearray; use decode_jpeg_header to find out required minimum size

• returns: image as numpy.ndarray

encode_jpeg

def encode_jpeg(
        image: numpy.ndarray,
        quality: SupportsInt=85,
        colorspace: Text='RGB',
        colorsubsampling: Text='444',
        fastdct: Any=True,
) -> bytes

Encode an image given as numpy array to JPEG (JFIF) string. Returns JPEG (JFIF) data.

• image: uncompressed image as uint8 array

• quality: JPEG quantization factor; 0-100, higher equals better quality

• colorspace: source colorspace; one of ‘RGB’, ‘BGR’, ‘RGBX’, ‘BGRX’, ‘XBGR’, ‘XRGB’, ‘GRAY’, ‘RGBA’, ‘BGRA’, ‘ABGR’, ‘ARGB’, ‘CMYK’

• colorsubsampling: subsampling factor for color channels; one of ‘444’, ‘422’, ‘420’, ‘440’, ‘411’, ‘Gray’.

• fastdct: If True, use fastest DCT method; usually no observable difference

• returns: bytes object of encoded image as JPEG (JFIF) data

is_jpeg

def is_jpeg(data: SupportsBytes)

Check whether a bytes object (or similar) contains JPEG (JFIF) data.

• data: JPEG (JFIF) data

• returns: True if JPEG