This tutorial shows you how to implement a simple JPEG encoder (sequential mode) step by step.
Implement the following parts
-
RGB to YUV color transform
-
Sample image blocks, you are required to implement this method for YUV 444, YUV 422, and YUV 420
For sub-sampling blocks, you can use nearestorbilinearinterpolation (e.g., directly average a 2x2 area into a new pixel for YUV420). -
DCT (you are not allowed to use external library such as cv2.dct)
the time complexity of the original dct is O(n^4), and there are many optimized tricks can be found online.
Once implemented the above functions, you can test the jpeg encoder with ``data/sg_0.png'' or other images. You are encouraged to test images with different quality and yuv format settings.
- Learn the pipeline and details of a JPEG encoder
- Implement a simple JPEG encoder for fun, e.g., given a PNG image with size over 30MB, with your JPEG encoder, it can be compressed to no more than 1MB.
git clone --recursive https://github.com/JosephChenHub/comp5425_assignment2_jpeg_encoder.git
We use ``cmake >= 3.13 " to build the project, actually you can use other compilers like Visual Studio, MingW, etc. 0. build
cd ./jpeg_encoder && \
mkdir build && cd build \
cmake .. \
make -j
use -DBUILD_TESTS=ON to enable tests, -DBUILD_PYTHON_MODULE=ON to build python interface.
usage example:
./build/jpeg_encoder -i ./data/sg_0.png -o ./data/sg_0_q30_420.jpg -q 30 -f 420
This command will generate a JPEG image with quality 30 and YUV420 format.
Some APIs of Image class:
Image<uint8_t> img1("./data/sg_0.png"); // load an image from file, only for uint8_t type
Image<uint8_t> img2 = img1; // copy
img2(0, 5, 2) = 10; // assign the pixel value to 10 at location(y=0, x=5, c=2)
Image<float> img3;
img3.copyFrom(img2); // uint8_t -> float copy
TASK : For cpp users, you are required to complete functions in src/JpegEncoder.cpp
Method 1 (via CMake)
The python version is based on cpp version (Huffman coding & JPEG IO), so you still need to build jpeg_py first: 0. add the pybind11 if you need a python wrapper
cd ./jpeg_encoder && \
git submodule add https://github.com/pybind/pybind11.git
- build the jpeg_py module:
cd ./jpeg_encoder && \
mkdir build && cd build \
cmake .. -DBUILD_PYTHON_MODULE=ON \
make -j
- add jpeg_py to the Python environment :
import sys
sys.path.append("./build") # the path -> jpeg_py module, you should first build the module via cmake first
from jpeg_py import YUVFormat, JpegIO, HuffmanCodec
Then you can use the following APIs:
- YUVFormat.YUV444 : YUV444 , i.e., no subsampling for luminance & chrominance
- YUVFormat.YUV420 : YUV420 , the Micro Block will look like [YYYYUV][YYYYUV]... [YYYYUV], i.e., subsampling chrominance from vertical (factor 2) & horizontal (factor 2) direction
- YUVFormat.YUV422 : YUV4222, subsampling chrominance from horizontal direction (factor 2), [YYYYUUVV][YYYYUUVV]...[YYYYUUVV]
- JpegIO.writeToFile: write the encoded image to disk
- HuffmanCodec : huffman coding for Y, U, V blocks
Method 2(via setup.py):
python3 setup.py build
then add the './build/xxx' to your python environment, or install the package via "python3 setup.py install"
TASK : For python users, you are required to complete functions in python/jpeg_encoder.py.
- COMP5425, HK PolyU (lecturer: Prof. Changwen Chen)
- https://en.wikipedia.org/wiki/JPEG
- https://yasoob.me/posts/understanding-and-writing-jpeg-decoder-in-python/#discrete-cosine-transform-quantization
- Digital Video Concepts, Methods, and Metrics: Quality, Compression, Performance, and Power Trade-off Analysis
- https://users.ece.utexas.edu/~ryerraballi/MSB/pdfs/M4L1.pdf
- https://en.wikipedia.org/wiki/Chroma_subsampling#Sampling_systems_and_ratios