Contour is one of the most important concept in plenty of image-based applications, mainly for the representation of region of interest (ROI), especially in medical imaging applications. OpenCV and shapely both provide some contour functionalities. However, the contour representation looks obscure and strange for developers.
This package tries to standardize the contour representation in python. Different from OpenCV, the contour here is represented as a numpy 2d array with shape 2*N (0-h 1-w), in which height always comes first, shown as the figure above. We believe this representation is the most intuitive way for developers to understand and use. Moreover, back and forth conversion between this representation with OpenCV or shapely are supported, enabling developers to take advantage of the functionalities of both OpenCV and shapely. We also add contour utilities, such as feature extraction, statistic metric calculation, shifting and rotation, mask construction, etc. to simplify contour usage. If you find pycontour to be helpful for your work, please star this repo.
To install pycontour, libgeos need to install in advance. Other required packages can refer to requirements.txt.
# install dependencies
sudo apt-get install libgeos-dev
# install relied packages
pip install -r requirements.txt
# install pycontour
pip install pycontour==1.5.1
# Representation Transformation
- back-and-forth transformation with cv2 representation
- back-and-forth transformation with shapely Polygon
```python
import numpy as np
# with OpenCV Representation
from pycontour.cv2_transform import cv_cnt_to_np_arr, np_arr_to_cv_cnt
np_arr1 = np.array([[1, 0, 2, 3, 5, 4], [0, 2, 4, 3, 1, 0]])
cv_cnt = np_arr_to_cv_cnt(np_arr1)
np_arr2 = cv_cnt_to_np_arr(cv_cnt)
assert np.array_equal(np_arr1, np_arr2), "Back and forth transformation not equal"
# with Shapely Polygon
from pycontour.poly_transform import np_arr_to_poly, poly_to_np_arr
np_arr1 = np.array([[1, 0, 2, 3, 5, 4], [0, 2, 4, 3, 1, 0]])
poly = np_arr_to_poly(np_arr1)
np_arr2 = poly_to_np_arr(poly)
assert np.array_equal(np_arr1, np_arr2), "Back and forth transformation not equal"
# with Point List
from pycontour.coor_transform import point_list_to_np_arr, np_arr_to_point_list
np_arr1 = np.array([[1, 0, 2, 3, 5, 4], [0, 2, 4, 3, 1, 0]])
point_list = np_arr_to_point_list(np_arr1)
np_arr2 = point_list_to_np_arr(point_list)
assert np.array_equal(np_arr1, np_arr2), "Back and forth transformation not equal"
```
# Contour Features
- area, aspect ratio, solidity, bounds, etc.
# Interconversion of Contour and Image
- constructing binary image based on contour
- masking image based on contour
```python
import numpy as np
from pycontour.img import build_cnt_mask
np_arr1 = np.array([[1, 0, 2, 3, 5, 4], [0, 2, 4, 3, 1, 0]])
mask = build_cnt_mask(np_arr1)
```
# Feature Extraction
- ZernikeMoments feature extraction
```python
import numpy as np
from pycontour.fea import ZernikeMoments
np_arr1 = np.array([[1, 0, 2, 3, 5, 4], [0, 2, 4, 3, 1, 0]])
zernike_desc = ZernikeMoments(radius=21)
cnt_fea = zernike_desc.cal_fea(np_arr1)
assert len(cnt_fea) == 25, "Feature error"
```
# Contour Relationship
- point with contour
- contour with contour
- overlapping ratio (Dice coefficient, Jaccard index, etc.)
# Shape Transformation
- rotate, shift, smooth, etc.
Hosted in https://pycontour.readthedocs.io, powered by readthedocs and Sphinx.
pycontour is an open source project and anyone is welcome to contribute. An easy way to get started is by suggesting a new enhancement on the Issues. If you have found a bug, then either report this through Issues, or even better, fork the repository, fix the bug and then create a Pull Request.
See the AUTHORS.md file for a complete list of contributors to the project.
pycontour is free software made available under the BSD 3-Clause License. For details see the LICENSE file.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent