ujpeg

Jpeg encoder/decoder library for micropython

Currently only decoding is implemented. Two functions are supported:

• decode from jpeg file to ndarray (ulab)
• decode from jpeg file to ppm file

Defaults in ulab (compilation time option) sets the maximum dimension of a ndarray to 2 dimension. Thus to stay compatible with default ulab compilations a color image will be returned as a tuple of three 2D ndarrays, one for each channel:

import ujpeg
from ulab import numpy as np

r, g, b = ujpeg.decode("test.jpg")
print("Image shape: %dx%d" % r.shape)

idx = np.argmax(list(map(np.mean, [r, g, b])))
colors = "red green blue".split()
print("Image is very %s!" % colors[idx])

An image can also be directly decoded to another image file:

import ujpeg

ujpeg.decode_to_file("test.jpg", "test.pgm")

Idea is to have a simple micropython library to

• encode jpeg images from raw bytes
• decode jpeg images from raw bytes
• decode jpeg image from file

How to use

The file structure follows the regular micropython c module workflow, see official instructions for details.

You need to link against libjpeg. If you are using Ubuntu or debian, do

apt-get install libjpeg-dev -y

or if cross-compiling, e.g.

apt-get install libjpeg-dev:armhf -y

A project using this specific library can be found here: https://github.com/daniel-falk/camera-analytics

Contribute

All contributions are welcomed, such as:

• patchsets, bug fixes
• documentation
• ideas of API or refactorization
• feature requests

The c-code test file can be compiled with:

cd src
gcc test_main.c jpeg_encode_decode.c -Wall -Werror -ljpeg
./a.out <path_to_file.jpg>

Python test file can be run after building micropython with this module:

./micropython test.py

Coding style

C code should follow the Chromium style but with 4 spaces indent. This can be automatically done using e.g. clang-format-9:

clang-format-9 -style="{BasedOnStyle: chromium,IndentWidth: 4}" -i *.[ch]

Finding areas of improvement

Start by reading the TODO labels in the code, e.g. using grep:

$ grep -nr "TODO"
jpeg_encode_decode.c:32:    // TODO: Risk for loop-for-ever if some error...
jpeg_encode_decode.c:49: * TODO: Add option to specify color space
jpeg_encode_decode.c:50: * TODO: Add option to scale image (libjpeg scale_num, scale_denom)
jpeg_encode_decode.c:51: * TODO: Add ability to reuse the same decompress object for e.g. video
jpeg_encode_decode.c:72:    // TODO: Replace error handler with something that won't exit..
jpeg_encode_decode.c:84:        // TODO: Seems like the return value should not be needed to check
jpeg_encode_decode.c:111:    // TODO: Use rec_outbuf_height lines... Avoid using too many at end
jpeg_encode_decode.c:123:    // TODO Remove this and return image data instead...