We implement a simple pb-lite edge detection algorithm in this project. First, generate four filter banks which are oriented Derivative of Gaussian (DoG), Leung-Malik (LM), Gabor, and Half-disc. By using these filters, we can create a texture map of the image. Also, use KMeans clustering to generate brightness and color. Then, implement chi-square distance combined with sobel and canny edges and will have pb-lite edges of the image. You can find the details of the project on this website.
The overview of the pb-lite edge detection algorithm is shown below.
- opencv-python
- numpy
- tqdm
- scikit-learn
- scikit-image
- matplotlib
BSDS500: Images of canny and Sobel baselines. Also with original images.Code: All codes for generating pb-lite edge detectionFigures: Some results of pb-lite edge detection algorithm
Use the following command to generate the results of the pb-lite edge detection
python3 ./Wrapper.py
The figures below show the color, brightness, texture maps, and their gradient maps. Also shows the comparison between canny, sobel, and pb-lite edge detection algorithms.
| Brightness Gradient | Color Gradient | Texture Gradient |
|---|---|---|
 |
 |
 |
| Brightness Map | Color Map | Texton Map |
|---|---|---|
 |
 |
 |
| Canny Baseline | Sobel Baseline | Pb-Lite |
|---|---|---|
 |
 |
 |
Arbelaez, Pablo, et al. "Contour detection and hierarchical image segmentation." IEEE transactions on pattern analysis and machine intelligence 33.5 (2010): 898-916.