-Initialize opencv rng by iterating generator random-n times prior to training data

-Import matlib library directly
-Changed replicates to 5
-Pass zero matrix to execute kmeans++ alg in opencv (OpenCV does not use  means that are passed in)

srcx/Python/.idea/Python.iml

@@ -2,7 +2,7 @@
 <module type="PYTHON_MODULE" version="4">
   <component name="NewModuleRootManager">
     <content url="file://$MODULE_DIR$" />
-    <orderEntry type="jdk" jdkName="Python 3.5.1 (C:\Users\angeline\Anaconda3\python.exe)" jdkType="Python SDK" />
+    <orderEntry type="inheritedJdk" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
   <component name="TestRunnerService">

srcx/Python/.idea/misc.xml

 <?xml version="1.0" encoding="UTF-8"?>
 <project version="4">
-  <component name="ProjectLevelVcsManager" settingsEditedManually="false">
-    <OptionsSetting value="true" id="Add" />
-    <OptionsSetting value="true" id="Remove" />
-    <OptionsSetting value="true" id="Checkout" />
-    <OptionsSetting value="true" id="Update" />
-    <OptionsSetting value="true" id="Status" />
-    <OptionsSetting value="true" id="Edit" />
-    <ConfirmationsSetting value="0" id="Add" />
-    <ConfirmationsSetting value="0" id="Remove" />
-  </component>
-  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.5.1 (C:\Users\angeline\Anaconda3\python.exe)" project-jdk-type="Python SDK" />
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.5.1 (C:\Users\angeline\Anaconda3\envs\AnacondaEnv\python.exe)" project-jdk-type="Python SDK" />
 </project>
\ No newline at end of file

srcx/Python/PIXEL_REPLICATION_LIB.py

@@ -37,6 +37,7 @@
 # respective depth of the pixel in the Euclidean distance transform.
 def PixelReplicate(bwim):
     import numpy as np
+    from numpy import matlib as matlib
     import cv2
 
     # Computer Euclidean Distance Transform
@@ -64,7 +65,7 @@ def PixelReplicate(bwim):
         nrep = np.round(bwd[idxPts[0, 0], idxPts[0, 1]])
         nrep = int(nrep)
         nrepTotal = np.append(nrepTotal,nrep)
-        rp = np.matlib.repmat(idxPts, nrep, 1)
+        rp = matlib.repmat(idxPts, nrep, 1)
         if (nInsert + nrep) > nPtsRep:
             ptsRep = np.append(ptsRep, rp, axis=0)
         else:
@@ -81,20 +82,20 @@ def PixelReplicate(bwim):
 def fitGMM(ptsRep, k):
     import numpy as np
     import cv2
+    import time
     print("Fitting GMM...")
     bestGMM = []
     bestLL = float("-inf") # initialize lowest possible log likelihood value (-inf)
 
-    # Get unique points from replicated points
-    coorPtsRep = [tuple(x) for x in ptsRep]
-    uniqPts = sorted(set(coorPtsRep), key=lambda x: coorPtsRep.index(x))
-    uniqPts = np.array(uniqPts)
+    # Initialize random number generator
+    n = int(np.round(time.time() % 100))
+    for i in np.arange(0,n,1):
+        cv2.randu(1, 1, 10)
 
     # GAUSSIAN MIXTURE MODEL FITTING
-    for i in np.arange(0, 10, 1): # Initialization Loop
-        # Random generation of indices into non-replicated points array
-        rndIdx = np.random.randint(low=0, high=uniqPts.shape[0], size=(1, 1, k), dtype=int)
-        rndMeans = np.matrix(uniqPts[np.array(rndIdx), 0:])
+    for i in np.arange(0, 5, 1): # Initialization Loop
+        # Must pass in zero matrix of means to execute kmeans++ alg in opencv
+        startMeans = np.matrix(np.zeros((k,2)))
 
         # Initialize GMM model
         gmm0 = cv2.ml.EM_create()
@@ -104,16 +105,16 @@ def fitGMM(ptsRep, k):
         gmm0.setTermCriteria(termCrit)
 
         # Train the GMM
-        newGMM = gmm0.trainE(samples=ptsRep, means0=rndMeans)
+        newGMM = gmm0.trainE(samples=ptsRep, means0=startMeans)
 
         # Calculate Log Likelihood of fit
         LL = sum(newGMM[1]) # sum of log likelihoods per sample
-        print('Likelihood Logarithm of GMM fit (Rep. ', i, ') = ', LL)
 
         # Select better fit using log likelihood comparison
         if LL > bestLL: # closer to zero Negative-LL is better
             bestGMM = gmm0
             bestLL = LL
+
     return bestGMM # return best gmm model (openCV EM class)
 
 

srcx/Python/PixelRep_Demo.py

@@ -36,14 +36,14 @@ import GENERATE_ELLIPSE_LIB as GenEllipse
 # Generate random ellipse image
 print("Generating Ellipse...")
 K = 4 # number of ellipse to fit
-bwim, trueEdges = GenEllipse.GetRandomEllipseImage(K)
+# bwim, trueEdges = GenEllipse.GetRandomEllipseImage(K)
 
 # # You may import an image for demo by uncommenting the following code:
-# import cv2
-# im = cv2.imread('U:\Angeline\PixelRep\Ellipse_k4.tif', cv2.CV_8UC1)
-# th, bwim = cv2.threshold(im, 0, 1, cv2.THRESH_BINARY)
-# trueEdges = [0, 0] # keep as [0, 0] if true edges are unknown.
-#                    # Include (r, c) array of boundary pixels if known
+import cv2
+im = cv2.imread('U:\Angeline\~CODE\PixelRep\Ellipse_k4.tif', cv2.CV_8UC1)
+th, bwim = cv2.threshold(im, 0, 1, cv2.THRESH_BINARY)
+trueEdges = [0, 0] # keep as [0, 0] if true edges are unknown.
+                   # Include (r, c) array of boundary pixels if known
 # # (Remember to modify K number of ellipses)
 
 # IMPORTANT NOTE: