notebook.community

Edit and run



In [5]:

    
# to make matplotlib figures nice
import matplotlib
%matplotlib inline
matplotlib.rcParams['savefig.dpi'] = 2 * matplotlib.rcParams['savefig.dpi']

Import vigra and graph submodule, and import numpy from vigra



In [10]:

    
import vigra
from vigra import graphs
import numpy
import pylab

Read in an image



In [7]:

    
# load image and convert to LAB
filepath = '12003.jpg'
img = vigra.impex.readImage(filepath)
imgLab = vigra.colors.transform_RGB2Lab(img)
vigra.imshow(img,show=False)
vigra.show()

Compute gradient magnitude on image which shape*2-1



In [8]:

    
# compute gradient on interpolated image
sigmaGradMag = 2.0 
imgLabBig = vigra.resize(imgLab, [imgLab.shape[0]*2-1, imgLab.shape[1]*2-1])
gradMag = vigra.filters.gaussianGradientMagnitude(imgLabBig, sigmaGradMag)
vigra.imshow(gradMag,show=False)
vigra.show()

Get oversegmentation



In [11]:

    
# get super-pixels with slic on LAB image
superpixelDiameter = 20 # super-pixel size
slicWeight = 10.0       # SLIC color - spatial weight
imgLab = vigra.colors.transform_RGB2Lab(img)
labels, nseg = vigra.analysis.slicSuperpixels(imgLab, slicWeight,
                                              superpixelDiameter)
labels = vigra.analysis.labelImage(labels)
# A random colormap for matplotlib
cmap = matplotlib.colors.ListedColormap ( numpy.random.rand ( 256,3))
pylab.imshow ( labels.squeeze().swapaxes(0,1), cmap = cmap)
pylab.show()

vigra.segShow(img,labels,alpha=0.0)
vigra.show()

Get a 2d grid graph



In [12]:

    
gridGraph = graphs.gridGraph(img.shape[0:2])
print(gridGraph)









    



Nodes: 154401 Edges: 308000 maxNodeId: 154400 maxEdgeId: 308801

Compute a region adjacency graph from oversegmentation labeling and grid graph



In [13]:

    
# get region adjacency graph from super-pixel labels
rag = graphs.regionAdjacencyGraph(gridGraph, labels)
print(rag)









    



Nodes: 764 Edges: 1963 maxNodeId: 764 maxEdgeId: 1962

Get edge weight for grid graph



In [14]:

    
gridGraphEdgeIndicator = graphs.edgeFeaturesFromInterpolatedImage(gridGraph,
                                                                  gradMag)

print(gridGraphEdgeIndicator.shape)









    



(481, 321, 2)

Get edge weight and node features accumulated from grid graph edge weights and node features



In [16]:

    
# accumulate edge weights grid graph edge weights
edgeWeights = rag.accumulateEdgeFeatures(gridGraphEdgeIndicator)

# accumulate node features from grid graph node map
# which is just a plain image (with channels)
nodeFeatures = rag.accumulateNodeFeatures(imgLab)

print(edgeWeights.shape)
print(nodeFeatures.shape)









    



(1963,)
(765, 3)



In [17]:

    
# do agglomerativeClustering
beta = 0.5              # node vs edge weight
nodeNumStop = 50        # desired num. nodes in result
labels = graphs.agglomerativeClustering(graph=rag, edgeWeights=edgeWeights,
                                        beta=beta, nodeFeatures=nodeFeatures,
                                        nodeNumStop=nodeNumStop)
# show result 
imgLabels =rag.projectLabelsToBaseGraph(labels)
cmap = matplotlib.colors.ListedColormap ( numpy.random.rand ( 256,3))
pylab.imshow ( imgLabels.squeeze().swapaxes(0,1), cmap = cmap)
pylab.show()

rag.show(img,labels)
vigra.show()









    



prepare 
get edge length
get node size
nodeLabels  (765,) uint32



In [ ]: