Segment labelling using connected component algorithms on binary (or boolean) images are used to determine connectivity of image pixels. TensorFlow Addons supports this task, however, it only considers 4-way connectivity (neighbours above, below, left, and right), unlike OpenCV, which supports both 4-way and 8-way connectivity options. To bridge this gap, this module employs TensorFlow Addons' connected components operation to perform segment labelling based on 8-way connectivity.
NOTE: Please refer to the behaviour of TensorFlow Addons' connected components for segment id assignment when sending images in a batch.
import cv2
import tensorflow as tf
import components
# get the image and convert to binary 1s and 0s
# image source: https://stackoverflow.com/questions/46441893/connected-component-labeling-in-python
img = cv2.imread('eGaIy.jpg', 0)
img = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY)[1]
# get segment ids
segment_ids = components.ConnectedComponents()(
tf.cast(tf.expand_dims(tf.convert_to_tensor(img), axis=0), tf.bool)
)
CPU: Intel(R) Core(TM) i7-9700K CPU @ 3.60GHz
GPU: GeForce RTX 2080 Ti
TensorFlow: 2.4.1
Python: 3.8.10
The python packages and their versions used during the development of this module can be found in requirements.txt.
OpenCV: 0.0248 seconds for 1000 reps of eGaIy.jpg
Tensorflow Addons: 0.9348 seconds for a batch of 1000 eGaIy.jpg
Tensorflow (DNS): 1.0359 seconds for a batch of 1000 eGaIy.jpg