Update edge.py

mahotas/edge.py

@@ -20,9 +20,10 @@
     [ 1,  2,  1]])/8.
 
 __all__ = [
-    'sobel',
-    'dog',
-    ]
+	'sobel',
+	'sujoy',
+	'dog',
+	]
 
 def sobel(img, just_filter=False):
     '''
@@ -69,6 +70,64 @@ def sobel(img, just_filter=False):
     thresh = 2*np.sqrt(filtered.mean())
     return mh.regmax(filtered) * (np.sqrt(filtered) > thresh)
 
+def sujoy(img, kernel_nhood=0, just_filter=False):
+
+	'''
+	edges = sujoy(img, kernel_nhood=0, just_filter=False)
+
+	Compute edges using Sujoy's algorithm
+
+	`edges` is a binary image of edges computed according to Sujoy's algorithm.
+	Paper link: https://www.ijert.org/research/a-better-first-derivative-approach-for-edge-detection-IJERTV2IS110616.pdf
+
+	Parameters
+	----------
+	img : Any 2D-ndarray
+	kernel_nhood : 0(default) or 1
+		if 0, kernel is based on 4-neighborhood
+		else , kernel is based on 8-neighborhood
+	just_filter : boolean, optional
+		If true, then return the result of filtering the image with the Sujoy's
+		filters, but do not threshold (default is False).
+
+	Returns
+	-------
+	edges : ndarray
+		Binary image of edges, unless `just_filter`, in which case it will be
+		an array of floating point values.
+	'''
+	# This is similar to 'sobel" implementation above
+	img = np.array(img, dtype=np.float)
+	if img.ndim != 2:
+		raise ValueError('mahotas.sujoy: Only available for 2-dimensional images')
+	img -= img.min()
+	ptp = img.ptp()
+	if ptp == 0:
+		return img
+	img /= ptp
+
+	if kernel_nhood:
+		krnl_h = np.array([[0,-1,-1,-1,0],[0,-1,-1,-1,0],[0,0,0,0,0],[0,1,1,1,0],[0,1,1,1,0]])/12.
+		krnl_v = np.array([[0,0,0,0,0],[-1,-1,0,1,1],[-1,-1,0,1,1],[-1,-1,0,1,1],[0,0,0,0,0]])/12.
+	else:
+		krnl_h = np.array([[0,0,-1,0,0],[0,-1,-1,-1,0],[0,0,0,0,0],[0,1,1,1,0],[0,0,1,0,0]])/8.
+		krnl_v = np.array([[0,0,0,0,0],[0,-1,0,1,0],[-1,-1,0,1,1],[0,-1,0,1,0],[0,0,0,0,0]])/8.
+
+	grad_h = convolve(img, krnl_h, mode='nearest')
+	grad_v = convolve(img, krnl_v, mode='nearest')
+
+	grad_h **=2
+	grad_v **=2
+
+	grad = grad_h
+	grad += grad_v
+	if just_filter:
+		return grad
+	t = np.sqrt(grad.mean())
+
+	return mh.regmax(grad)*(np.sqrt(grad)>t)
+
+
 def dog(img, sigma1 = 2, multiplier = 1.001, just_filter = False):
     '''
     edges = dog(img, sigma1 = 2, thresh= None, just_filter = False)