Use pip to install needed libraries.
pip install numpy matplotlib PillowPython3+ is needed
After running the file "k-compress.py", the porgram will need to know the name of the file and the k-value/number of colors needed.
Filename: king.png
K_value: 2! PS: The imagen must be in the folder img/original AND be in the format PNG !
Imagenes contains building-block known as pixels. A pixel is an entity that has three values; Red, Green and Blue. Each value of RGB ranges from 0-255 going from not containing any of the colours or containing maximum(255). In other words, a green, a white and a magenta pixel has the values in respective order;
- Green = rgb(0, 255, 0)
- White = rgb(255, 255, 255)
- Magenta = rgb(255, 0, 255)
The main algorithms making all the magic happening is the k-means clustering [https://en.wikipedia.org/wiki/K-means_clustering].
As previously explained, the pixel contains three values. The three values can be used as axes creating a three-dimensional plane. Every pixel in the imagen is inserted in the plane, giving it a position using its RGB values. The usage of the algorithm is to find clusterings of a set of, in this case, points. The algorithm needs a K-values indicating how many clusters that are wanted. In this context, the founded points after running the algorithm, are also a pixel with RGB values. Therefore, the k value in this context will be the number of wanted colours for the output imagen. Now we have K-points for all the pixels of the imagen.
The imagen can now be recreated by assigning every pixel the same values as the pixel's belonging clustering value.
Initially, the idea was to use Huffman-code to create an image, but as I found out, the PNG format already has a form of compression[https://en.wikipedia.org/wiki/Portable_Network_Graphics#Compression].
As the example imagen called "king.png" in the folder "img/original" of the Norwegian King Harald V, the original imagen has the size of 1 001 305 bytes. After converting it only to contain two colours, the size changed to 25 524 bytes and reduced in size by 97.46 %
| K (colours) | 2 | 4 | 8 | 16 | 32 | 64 | 128 | 256 |
|---|---|---|---|---|---|---|---|---|
| Size(bytes) | 25 524 | 52 090 | 148 847 | 209 680 | 360 498 | 509 475 | 701 327 | 900 328 |
| Reduction | 97.46% | 94.80% | 85.13% | 79.05% | 64% | 49.12% | 29.95% | 10.08% |
Here are the corresponding images. The K number increases to the power of 2, and correspond to the number of different colours.
