Extremly fast, (mostly) lossless JPEG transformations
A Python package for blazingly fast JPEG transformations. Compared to other, more general purpose image processing libraries like wand-py or PIL/Pillow, the performance gain can, depending on the transformation, be somewhere in the range of 150% to 500% (see Benchmarks). In addition to that, all operations except for scaling are lossless, since the image is not being re-compressed in the process. This is due to the fact that all transformation operations work directly with the JPEG data.
This is achieved by using multiple C routines from Carsten Haitzler’s epeg (scaling) and jpegtran from the Independent JPEG Group’s libjpeg library (for all other operations). These routines are called from Python through the CFFI library, i.e. no external commands are launched via subprocess.
- libjpeg8 with headers (earlier versions will not work)
$ pip install jpegtran-cffi
$ pip install cffi # cffi has to be installed before running setup.py $ python setup.py install
Before each transformation, you have to create a JPEGImage object. This can be either initialized with a filename (fname) or a bytestring with the JPEG data (blob).
On the resulting object, the following transformations are supported:
- Rotate by -90, 90, 180 or 270 degrees
- Flip either in vertical or horizontal direction
- “Tranpose” the image
- “Transverse” the image
- crop(x, y, width, height)
- Crop a rectangle with width and height starting from x pixels on the right and y pixels from the top
- scale(width, height, quality=75)
- Resize the image to width by height pixels. Since this is a lossy operation, the optional quality parameter can be used to set the JPEG quality of the output
The result of each of these operations retuns the same object with the new, transformed data. You can get the data as a bytestring via the as_blob() method or save it to a file with the save(fname) method.
Note that due to this layout, operations can be chained, e.g. to create a rectangular 200x200 thumbnail from a 1200x2400 pixel JPEG:
thumb_data = (JPEGImage(fname='original.jpg') .scale(200, 400) .crop(0, 100, 200, 200) .as_blob())
Wand-Py Image.sample(200, 150), filtering was nearest neighbour
PIL Image.resize((200, 150))
jpegtran-cffi JPEGImage.scale(200, 150, quality=75)
All operations were done on the following 2560x1920 8bit RGB JPEG: http://upload.wikimedia.org/wikipedia/commons/8/82/Mandel_zoom_05_tail_part.jpg
- CPU: 4x email@example.comGHz
- RAM: 16GB
- Storage: 7200rpm HDD
- PIL/Pillow: 2.3.0
- wand-py: 0.3.5
- jpegtran-cffi: 0.1
|scale to 250x150||102ms/299%||90ms/262%||34ms/0%|
|rotate by 90° CW||317ms/224%||258ms/182%||141ms/0%|
|Crop 500x500 from 0,0||190ms/475%||92ms/230%||40ms/0%|
Both wand-py and PIL were run with the fastest scaling algorithm available, for wand-py this meant using Image.sample instead of Image.resize and for PIL the nearest-neighbour filter was used for the Image.resize call.
The MIT License (MIT)
Copyright (c) 2014 Johannes Baiter <firstname.lastname@example.org>
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE