This is a utility to draw a set of points and to generate an image as if it was hand drawn by a human.
The need of having this component comes from an IA application, where a drawing is captured from a hand device (i.e. tablet or smartphone), and then the image is classified using a Neural Network. The points are captured to show the drawing to a human, who makes the initial classification to train the NN.
The effect is shown in the next image:
And using the same input data, it is possible to obtain different resolutions, colors, etc. by varying the settings:
Most of the code of this library is a port to python of library canvashanddraw.
Technical facts
- Includes tools to optimize the number of points, by skipping those that are too near to the line described by the adjacents points.
- It is possible to simulate the drawing dynamics by considering that the distance between the points is related to the speed of drawing. In that case, the faster a line is, the thinner it will be drawn, to simulate a ball pen.
The easier method to install canvashanddraw
is using pip:
pip install canvashanddraw
The alternate method is to clone the repo, install the dependencies and install by hand
$ git clone https://github.com/dealfonso/pycanvashanddraw
$ cd pycanvashanddraw
$ pip install pillow imageio aggdraw
$ python setup.py install
The dependencies from canvashanddraw
are installed when installing from pip, but you can also install them by hand:
pip install pillow imageio aggdraw
WARNING: canvashanddraw
depends on aggdraw
library, which is also in pip (so it is automatically installed when installing from pip). The problem is that at the time of writing this text, the current version of that library does not produce the best results. Instead, my advise is to remove the version installed using pip and to install the latest aggdraw
by hand:
$ pip uninstall aggdraw
...
$ mkdir -p ./tmp
$ cd ./tmp
$ git clone https://github.com/pytroll/aggdraw
$ cd aggdraw
$ python setup.py build_ext -i
$ python setup.py install
Canvas Hand Draw is conceived both as a command line utility and a library so that you can include it in your application.
### Command Line Utility
If you have a set of points in a json file (e.g. file "five.json")
five.json
[{"x": 211.33, "y": 130}, {"x": 210, "y": 130}, {"x": 199.33, "y": 130}, {"x": 177.66, "y": 130}, {"x": 154.33, "y": 130}, {"x": 135.33, "y": 130.66}, {"x": 124.33, "y": 132.33}, {"x": 120, "y": 133}, {"x": 118.66, "y": 133.33}, {"x": 118.66, "y": 133.66}, {"x": 118.66, "y": 134}, {"x": 118.66, "y": 138.66}, {"x": 118.66, "y": 144.33}, {"x": 118.33, "y": 159.33}, {"x": 116.66, "y": 176.66}, {"x": 114.66, "y": 191.33}, {"x": 113.33, "y": 201.33}, {"x": 112.66, "y": 206.33}, {"x": 112.33, "y": 208.33}, {"x": 112.33, "y": 209.33}, {"x": 112.33, "y": 210}, {"x": 113, "y": 209.66}, {"x": 118, "y": 208.66}, {"x": 131.33, "y": 205.66}, {"x": 150.66, "y": 202.33}, {"x": 171, "y": 200.33}, {"x": 186.66, "y": 200}, {"x": 197.33, "y": 200.66}, {"x": 205.33, "y": 203.66}, {"x": 210.33, "y": 207.33}, {"x": 213, "y": 212}, {"x": 213.66, "y": 219.66}, {"x": 212, "y": 230.33}, {"x": 208.33, "y": 241}, {"x": 202.66, "y": 251}, {"x": 195.66, "y": 260.66}, {"x": 183.33, "y": 272.33}, {"x": 173, "y": 279.66}, {"x": 165.33, "y": 283.33}, {"x": 150, "y": 289}, {"x": 137, "y": 292.33}, {"x": 127.66, "y": 294.66}, {"x": 122, "y": 295.66}, {"x": 119.33, "y": 296}, {"x": 119, "y": 296}]
You can run the next command line:
$ canvashanddraw -i five.json -o five.png -w 64
To get the next drawing:
Having the same file five.json
, you can use the next piece of code
from canvashanddraw.points import Points
from canvashanddraw.drawhelper import DrawHelper, Options
import json
points = json.load(open('five.json'))
points = Points(points)
helper = DrawHelper(points, Options(canvasSize = 64, canvasMargin = 4, lineWidth = 4, lineColor = "#f00"))
helper.draw().save('five-red.png')
To generate the next drawing:
The points used to draw are defined in user-scale and they are later scaled to the target output.
Moreover, a single draw may consist of multiple drawings. In such case, the array of points used to define the drawing should include a null
or an empty point (e.g. []
) to separate the drawings:
[{"x": 211.33, "y": 130}, {"x": 210, "y": 130}, {"x": 199.33, "y": 130}, {"x": 177.66, "y": 130}, null, {"x": 135.33, "y": 130.66}, {"x": 124.33, "y": 132.33}, {"x": 120, "y": 133}, {"x": 118.66, "y": 133.33}, {"x": 118.66, "y": 133.66}, [], {"x": 165.33, "y": 283.33}, {"x": 150, "y": 289}, {"x": 137, "y": 292.33}, {"x": 127.66, "y": 294.66}, {"x": 122, "y": 295.66}, {"x": 119.33, "y": 296}, {"x": 119, "y": 296}]
There are multiple options to customize your results: the line width, the target resolution, the color, etc.
If using the CLI app, you are invited to use option --help
to get the information about the different options.
When using the library, please check class canvashanddraw.drawhelper.Options
:
class Options:
"""A class to represent the options for drawing"""
canvasSize = 64
canvasMargin = 4
lineWidth = 4
lineColor = '#000000'
drawDynamics = True
antiAliasing = True
keepAspectRatio = True