def computeSSF(img_name, filterSize=10, display=False, writeImage=False, verbose=False):
# read image
print "%s" % img_name
src = cv2.imread(img_name, 0).astype(np.uint32)
org = cv2.imread(img_name, 1)
gray = cv2.imread(img_name, 0)
# Area filter
if verbose:
print "Area Filter"
AF_img = tos.areaFilter(src, size=filterSize)
# Constructing tree of shapes
if verbose:
print "Constructing tree of shapes"
padding_img = tos.imagePadding(AF_img, 0)
tree = tos.constructTreeOfShapes(padding_img, None)
# attribute
if verbose:
print "Adding area & depth attributes"
tos.addAttributeArea(tree)
tos.addAttributeDepth(tree)
# generate graph and node position using NetworkX
if verbose:
print "Create Graph"
g, nodePixMap = createAttributeGraph(tree, padding_img)
# compute blobs and texture features
if verbose:
print "Compute blobs & features"
dark, bright, fVector = drawBlobs(org, tree, g, nodePixMap, gray)
# output
name, ext = os.path.splitext(img_name)
outName = name + ".csv"
np.savetxt(outName, [fVector], delimiter=",")
if verbose:
print "%s; done." % img_name
# write blob images
if writeImage:
cv2.imwrite("%s_dark.png" % name, dark)
cv2.imwrite("%s_bright.png" % name, bright)
# display
if display:
showImage(src, "src")
showImage(AF_img, "area filter 1")
cv2.imshow("dark_blobs", dark)
cv2.imshow("bright_blobs", bright)
cv2.waitKey()
return tree, g, nodePixMap
def SitaElements(filename, filterSize=10, fcomb='SI'):
"""
This function outputs two numpy array objects:
1. SI or AI
2. Normalized gray level values (without generating CtH)
fcomb: 'SI' or 'AI'
SI = EH + CpH + SRH
AI = EH + CpH
"""
src = cv2.imread(filename, 0).astype(np.uint32)
# constructing tree of shapes
AF_img = tos.areaFilter(src, size=filterSize)
padding = tos.imagePadding(AF_img, 0)
tree = tos.constructTreeOfShapes(padding, None)
tos.addAttributeArea(tree)
g, nodePixMap = createAttributeGraph(tree, padding)
# remove padding (root) node
removeEmptyRootNode(g, nodePixMap)
# compute each attribute histogram
EH, CpH, SRH, NL = computeSitaElementsFromTree(g, nodePixMap, AF_img)
# selecting output feature
if fcomb == 'AI':
feature = np.r_[EH, CpH]
elif fcomb == 'SI':
feature = np.r_[EH, CpH, SRH]
else:
print "Wrong string is specified. 'SI' was selected."
feature = np.r_[EH, CpH, SRH]
# output
basename, ext = os.path.splitext( filename )
if fcomb == 'AI':
np.save( basename + "-AI.npy", feature )
elif fcomb == 'SI':
np.save( basename + "-SI.npy", feature )
else:
print "Wrong string is specified. 'SI' was selected."
np.save( basename + "-SI.npy", feature )
np.save( basename + "-NL.npy", NL )
def SITA(filename, filterSize=10, fcomb='SI', isCtH=True):
"""
fcomb: 'SI' or 'AI'
SI = EH + CpH + SRH
AI = EH + CpH
isCtH: boolian
If isCtH is True, CtH is concatenated to SI or AI.
"""
src = cv2.imread(filename, 0).astype(np.uint32)
# constructing tree of shapes
AF_img = tos.areaFilter(src, size=filterSize)
padding = tos.imagePadding(AF_img, 0)
tree = tos.constructTreeOfShapes(padding, None)
tos.addAttributeArea(tree)
g, nodePixMap = createAttributeGraph(tree, padding)
# remove paddint (root) node
removeEmptyRootNode(g, nodePixMap)
# compute each attribute histogram
EH, CpH, SRH, NL = computeSitaElementsFromTree(g, nodePixMap, AF_img)
# selecting output feature
if fcomb == 'AI':
feature = np.r_[EH, CpH]
elif fcomb == 'SI':
feature = np.r_[EH, CpH, SRH]
else:
print "Wrong string is specified. 'SI' was selected."
feature = np.r_[EH, CpH, SRH]
if isCtH:
CtH = np.histogram(NL, bins=50, range=(NL_min, NL_max), normed=False)[0]
if np.sum(CtH) == 0:
CtH = np.zeros(50, dtype='float')
else:
CtH = CtH / float( np.linalg.norm(CtH, ord=1) )
feature = np.r_[feature, CtH]
# output
basename, ext = os.path.splitext( filename )
np.save( basename + "-SITA.npy", feature )