python-xdog

Simple implementation of XDoG

There are two principle files: main.py to run xdog implementation with image and video_jit to run in "real time" with webcam

a) Original	b) Grayscale	c) Xdog GaryGrossi
d) Xdog Test Nat	e) Xdog Test	f) Xdog Hatch

---

Parameters

• a) Original image RGB
• b) Load as grayscale (OpenCV)
• c) XdogGary: sigma=0.5, k=200, gamma=0.98, epsilon=0.1, phi=10
• d) Xdog this project: sigma=1, k=1.6, gamma=0.5, epsilon=-0.5, phi=10
• e) Xdog this project: sigma=0.4, k=1.6, gamma=0.5, epsilon=-0.5, phi=10
• f) Xdog this project: hatchBlend function

---

Demo of video_jit.py

---

Dependencies

• Python 2.7

• OpenCV (Tested on 2.4.13)

   sudo apt-get install python-opencv
  

• Numpy 1.11.2

   pip install numpy
  

If you want to run video_jit.py:

• Numba

   conda install numba
  

There is a requirement.txt to install numpy and numba:

pip install -r requirements.txt

---

Test time in Difference of Gaussians (DoG) implementation

start = time.time()
img1 = cv2.GaussianBlur(img,(0,0),0.5)
elapsed = time.time() - start
elapsed = 0.000515937805176

start = time.time()	
sigma = 0.5
gauss1 = gaussian_filter(img, 0.5)
elapsed = time.time() - start
print elapsed

elapsed = 0.00245594978333

So opencv cv2.GaussianBlur is faster. The difference from gaussian blur results from OpenCV and SciPy is similar.
The difference from each mat can be seen in image below:

References used in these project

Xdog Article

garygrossi/XDoG-Python