def computeTrainingData(args):
#A list of the different file suffixes corresponding to the different face variants for each subject
name = ['9326871','9336923','ahodki','anpage','doraj','ekavaz','jmedin','klclar','mbutle','moors','obeidn','pspliu']
#For each person
for i in name:
imageStorage = []
#We use the first 8 of 11 images as training data, and the final 3 as tests
for j in range(1,7):
img = dia.openImage('TrainingData/' + i + '.' + str(j) + '.jpg')[0]
imageStorage.append(img)
imageCube = np.array(imageStorage)
PCImages = PCATrans(imageCube)
#Save the component images to form the database of training data
for l in range(len(PCImages)):
dia.saveImage(np.uint8(255.0*(PCImages[l]-np.amin(PCImages[l]))/(np.amax(PCImages[l])-np.amin(PCImages[l]))), 'TrainingData/PrincipalComponents/' + i + '.' + str(l) + '.jpg')
np.savetxt('TrainingData/PrincipalComponents/' + i + '.component' + str(l) + '.txt',PCImages[l])
mean = np.sum(imageCube, axis=0)/(imageCube.shape[0])
dia.saveImage(np.uint8(255.0*(mean - np.amin(mean))/(np.amax(mean) - np.amin(mean))), 'TrainingData/PrincipalComponents/' + str(i) + 'mean.jpg')
mean = mean/np.sqrt(np.sum(mean*mean))
np.savetxt('TrainingData/PrincipalComponents/' + i + 'mean.txt',mean)
def computeMeanValues(args):
outfile = open("ImageMeanValues1.txt", 'w')
goodMeans = []
yellowedMeans = []
for i in range(3118,4116):
image = []
try:
if i < 10:
image = dia.openImage("DJI0000" + str(i) + ".JPG")
print "DJI0000" + str(i) + ".JPG"
elif i >= 10 and i < 100:
image = dia.openImage("DJI000" + str(i) + ".JPG")
print "DJI000" + str(i) + ".JPG"
elif i >= 100 and i < 1000 :
image = dia.openImage("DJI00" + str(i) + ".JPG")
print "DJI00" + str(i) + ".JPG"
else:
image = dia.openImage("DJI0" + str(i) + ".JPG")
print "DJI0" + str(i) + ".JPG"
meanVector = dia.meanVector(image)
if meanVector[2] > 80:
goodMeans.append(meanVector[2])
else:
yellowedMeans.append(meanVector[2])
outfile.write("Image Name: DJI00" + str(i) + ".JPG\n")
outfile.write("Image Mean Vector\n")
outfile.write(str(meanVector))
outfile.write('\n\n')
except AttributeError:
print "DJI00" + str(i) + ".JPG not found"
goodMeans = np.array(goodMeans, dtype='uint64')
yellowedMeans = np.array(yellowedMeans, dtype='uint64')
goodMeanGlobal = np.sum(1.0*goodMeans/goodMeans.shape[0])
yellowedMeanGlobal = np.sum(1.0*yellowedMeans/yellowedMeans.shape[0])
def computeTrainingData(args):
#A list of the different file suffixes corresponding to the different face variants for each subject
name = [
'9326871', '9336923', 'ahodki', 'anpage', 'doraj', 'ekavaz', 'jmedin',
'klclar', 'mbutle', 'moors', 'obeidn', 'pspliu'
]
#For each person
for i in name:
imageStorage = []
#We use the first 8 of 11 images as training data, and the final 3 as tests
for j in range(1, 7):
img = dia.openImage('TrainingData/' + i + '.' + str(j) + '.jpg')[0]
imageStorage.append(img)
imageCube = np.array(imageStorage)
PCImages = PCATrans(imageCube)
#Save the component images to form the database of training data
for l in range(len(PCImages)):
dia.saveImage(
np.uint8(255.0 * (PCImages[l] - np.amin(PCImages[l])) /
(np.amax(PCImages[l]) - np.amin(PCImages[l]))),
'TrainingData/PrincipalComponents/' + i + '.' + str(l) +
'.jpg')
np.savetxt(
'TrainingData/PrincipalComponents/' + i + '.component' +
str(l) + '.txt', PCImages[l])
mean = np.sum(imageCube, axis=0) / (imageCube.shape[0])
dia.saveImage(
np.uint8(255.0 * (mean - np.amin(mean)) /
(np.amax(mean) - np.amin(mean))),
'TrainingData/PrincipalComponents/' + str(i) + 'mean.jpg')
mean = mean / np.sqrt(np.sum(mean * mean))
np.savetxt('TrainingData/PrincipalComponents/' + i + 'mean.txt', mean)
def PCATrans(img):
mean = np.sum(img, axis=0) / (img.shape[0])
resImage = 1.0 * img - 1.0 * mean
imgCov = dia.varCorrMatrix(img)[0]
#Create a list containing the number of faces
bands = []
for i in range(img.shape[0]):
bands.append(i + 1)
#Compute the eigenvalues and eigenvectors of the covariance matrix
covEigVal, covEigVec = np.linalg.eig(imgCov)
#Output the eigenvalues, eigenvectors and %variance values to the output file
#outfile.write("EigenValues:\n" + str(covEigVal))
#outfile.write("\n\n%Variance:\n" + str(covEigVal/np.sum(covEigVal)))
#outfile.write("\n\nEigenVectors:\n" + str(covEigVec))
#Create an array of eigenvalue, eigenvector and band number pairs
pairs = [(covEigVal[i], covEigVec[i], bands[i]) for i in range(len(bands))]
#Sort the eigenvalue pairs in descending order
pairs.sort()
pairs.reverse()
#Generate the eigenvector transformation matrix (G)
G = []
for i in range(len(pairs)):
G.append(pairs[i][1])
#Convert the images into a linear image
linearImage = resImage.reshape(img.shape[0], img.shape[1] * img.shape[2])
G = np.array(G)
#Compute the principle components of the image
linearPC = np.dot(G.T, linearImage)
PC = []
#Transform from a linear image to a square image
for i in range(len(linearPC)):
temp = linearPC[i] / np.sqrt(np.sum(linearPC[i] * linearPC[i]))
PC.append(temp.reshape(img.shape[1], img.shape[2]))
return np.array(PC)
def PCATrans(img):
mean = np.sum(img, axis=0)/(img.shape[0])
resImage = 1.0*img - 1.0*mean
imgCov = dia.varCorrMatrix(img)[0]
#Create a list containing the number of faces
bands = []
for i in range(img.shape[0]):
bands.append(i + 1)
#Compute the eigenvalues and eigenvectors of the covariance matrix
covEigVal, covEigVec = np.linalg.eig(imgCov)
#Output the eigenvalues, eigenvectors and %variance values to the output file
#outfile.write("EigenValues:\n" + str(covEigVal))
#outfile.write("\n\n%Variance:\n" + str(covEigVal/np.sum(covEigVal)))
#outfile.write("\n\nEigenVectors:\n" + str(covEigVec))
#Create an array of eigenvalue, eigenvector and band number pairs
pairs = [(covEigVal[i],covEigVec[i],bands[i]) for i in range(len(bands))]
#Sort the eigenvalue pairs in descending order
pairs.sort()
pairs.reverse()
#Generate the eigenvector transformation matrix (G)
G = []
for i in range(len(pairs)):
G.append(pairs[i][1])
#Convert the images into a linear image
linearImage = resImage.reshape(img.shape[0],img.shape[1]*img.shape[2])
G = np.array(G)
#Compute the principle components of the image
linearPC = np.dot(G.T,linearImage)
PC = []
#Transform from a linear image to a square image
for i in range(len(linearPC)):
temp = linearPC[i]/np.sqrt(np.sum(linearPC[i]*linearPC[i]))
PC.append(temp.reshape(img.shape[1],img.shape[2]))
return np.array(PC)
def computeWeights():
#A list of the different file suffixes corresponding to the different face variants for each subject
name = [
'9326871', '9336923', 'ahodki', 'anpage', 'doraj', 'ekavaz', 'jmedin',
'klclar', 'mbutle', 'moors', 'obeidn', 'pspliu'
]
weightDB = {}
#For each person
for i in name:
imageStorage = []
weights = []
for j in range(1, 5):
img = np.loadtxt('TrainingData/PrincipalComponents/' + i +
'.component' + str(j) + '.txt')
imageStorage.append(img)
imageStorage = np.array(imageStorage)
mean = np.loadtxt('TrainingData/PrincipalComponents/' + i + 'mean.txt')
for j in range(1, 4):
img = dia.openImage('TrainingData/' + i + '.' + str(j) + '.jpg')[0]
img = img / np.sqrt(np.sum(img * img))
temp = []
for k in range(imageStorage.shape[0]):
temp.append(imageStorage[k] * (img - mean))
weights.append(np.array(temp))
weights = np.array(weights)
weightDB[i] = np.sum(weights, axis=0) / weights.shape[0]
##np.savetxt('RegularPrincipalComponents/subject' + str(i) + 'weights.txt',avgWeight)
return weightDB
def computeWeights():
#A list of the different file suffixes corresponding to the different face variants for each subject
name = ['9326871','9336923','ahodki','anpage','doraj','ekavaz','jmedin','klclar','mbutle','moors','obeidn','pspliu']
weightDB = {}
#For each person
for i in name:
imageStorage = []
weights = []
for j in range(1,5):
img = np.loadtxt('TrainingData/PrincipalComponents/' + i + '.component' + str(j) + '.txt')
imageStorage.append(img)
imageStorage = np.array(imageStorage)
mean = np.loadtxt('TrainingData/PrincipalComponents/' + i + 'mean.txt')
for j in range(1,4):
img = dia.openImage('TrainingData/' + i + '.' + str(j) + '.jpg')[0]
img = img/np.sqrt(np.sum(img*img))
temp = []
for k in range(imageStorage.shape[0]):
temp.append(imageStorage[k]*(img-mean))
weights.append(np.array(temp))
weights = np.array(weights)
weightDB[i] = np.sum(weights, axis=0)/weights.shape[0]
##np.savetxt('RegularPrincipalComponents/subject' + str(i) + 'weights.txt',avgWeight)
return weightDB
'9326871', '9336923', 'ahodki', 'anpage', 'doraj', 'ekavaz', 'jmedin',
'klclar', 'mbutle', 'moors', 'obeidn', 'pspliu'
]
computeTrainingData(sys.argv)
weightDB = computeWeights()
#For each person
summ = 0
total = 0
for i in name:
for j in range(7, 9):
image = dia.openImage('TrainingData/' + i + '.' + str(j) +
'.jpg')[0]
face = computeFace(image, weightDB, name)
print 'Image Name: subject ' + i + '.' + str(
j) + '\nIdentified as: subject0' + str(face) + '\n\n\n'
#input()
total += 1
if face == i:
summ += 1
print 1.0 * summ / total
if __name__=='__main__':
#A list of the different file suffixes corresponding to the different face variants for each subject
name = ['9326871','9336923','ahodki','anpage','doraj','ekavaz','jmedin','klclar','mbutle','moors','obeidn','pspliu']
computeTrainingData(sys.argv)
weightDB = computeWeights()
#For each person
summ = 0
total = 0
for i in name:
for j in range(7,9):
image = dia.openImage('TrainingData/' + i + '.' + str(j) + '.jpg')[0]
face = computeFace(image, weightDB, name)
print 'Image Name: subject ' + i + '.' + str(j) + '\nIdentified as: subject0' + str(face) + '\n\n\n'
#input()
total += 1
if face == i:
summ += 1
print 1.0*summ/total
def shiftImages(args):
outfile = open("ImageMeanValues1.txt", 'a')
yellowedMeans = []
goodMeans = []
redMeans = []
redGoodMeans = []
means = []
for i in range(3118,4116):
image = []
try:
if i < 10:
image = dia.openImage("DJI0000" + str(i) + ".JPG")
print "DJI0000" + str(i) + ".JPG"
elif i >= 10 and i < 100:
image = dia.openImage("DJI000" + str(i) + ".JPG")
print "DJI000" + str(i) + ".JPG"
elif i >= 100 and i < 1000 :
image = dia.openImage("DJI00" + str(i) + ".JPG")
print "DJI00" + str(i) + ".JPG"
else:
image = dia.openImage("DJI0" + str(i) + ".JPG")
print "DJI0" + str(i) + ".JPG"
meanVector = dia.meanVector(image)
means.append(meanVector)
if meanVector[2] > 80:
goodMeans.append(meanVector[2])
else:
yellowedMeans.append(meanVector[2])
if meanVector[0] < 145:
redGoodMeans.append(meanVector[0])
else:
redMeans.append(meanVector[0])
except AttributeError:
print "DJI00" + str(i) + ".JPG not found"
means = np.array(means)
plt.plot(range(0,means.shape[0]), means[:,0], c='red')
plt.plot(range(0,means.shape[0]), means[:,1], c='green')
plt.plot(range(0,means.shape[0]), means[:,2], c='blue')
plt.title('Image Mean Values')
plt.xlabel('Image Number')
plt.ylabel('Mean DN')
plt.show()
goodMeans = np.array(goodMeans, dtype='uint64')
yellowedMeans = np.array(yellowedMeans, dtype='uint64')
goodMeanGlobal = np.sum(1.0*goodMeans/goodMeans.shape[0])
yellowedMeanGlobal = np.sum(1.0*yellowedMeans/yellowedMeans.shape[0])
meanDifference = goodMeanGlobal - yellowedMeanGlobal
outfile.write("\n\n\n\n These are the global statistics of the blue band using a value of 80 to separate the \"good images\" from the \"yellowed images\". The value was obtained by inspection of the output mean values.\n\n")
outfile.write('\nGood Mean = ' + str(goodMeanGlobal))
outfile.write('\nYellow Mean = ' + str(yellowedMeanGlobal))
goodVar = np.sum(np.power(goodMeans - goodMeanGlobal,2))/(goodMeans.shape[0] - 1)
yellowedVar = np.sum(np.power(yellowedMeans - yellowedMeanGlobal,2))/(yellowedMeans.shape[0] - 1)
outfile.write('\nGood Standard Deviation = ' + str(np.sqrt(goodVar)))
outfile.write('\nYellowed Standard Deviaiton = ' + str(np.sqrt(yellowedVar)))
outfile.write('\nYellowed Range = ' + str(np.amax(yellowedMeans) - np.amin(yellowedMeans)))
outfile.write('\nGood Range = ' + str(np.amax(goodMeans) - np.amin(goodMeans)))
redGoodMeans = np.array(redGoodMeans, dtype='uint64')
redMeans = np.array(redMeans, dtype='uint64')
redGoodMeanGlobal = np.sum(1.0*redGoodMeans/redGoodMeans.shape[0])
redMeanGlobal = np.sum(1.0*redMeans/redMeans.shape[0])
redMeanDifference = redGoodMeanGlobal - redMeanGlobal
outfile.write("\n\n\n\n These are the global statistics of the Red band using a value of 145 to separate the \"good images\" from the \"yellowed images\". The value was obtained by inspection of the output mean values.\n\n")
outfile.write('\nGood Mean = ' + str(redGoodMeanGlobal))
outfile.write('\nYellow Mean = ' + str(redMeanGlobal))
goodVar = np.sum(np.power(redGoodMeans - redGoodMeanGlobal,2))/(redGoodMeans.shape[0] - 1)
yellowedVar = np.sum(np.power(redMeans - redMeanGlobal,2))/(redMeans.shape[0] - 1)
outfile.write('\nGood Standard Deviation = ' + str(np.sqrt(goodVar)))
outfile.write('\nYellowed Standard Deviaiton = ' + str(np.sqrt(yellowedVar)))
outfile.write('\nYellowed Range = ' + str(np.amax(redMeans) - np.amin(redMeans)))
outfile.write('\nGood Range = ' + str(np.amax(redGoodMeans) - np.amin(redGoodMeans)))
refImage = dia.openImage("DJI03790.JPG")
for i in range(3118,4116):
image = []
try:
if i < 10:
image = dia.openImage("DJI0000" + str(i) + ".JPG")
print "DJI0000" + str(i) + ".JPG"
elif i >= 10 and i < 100:
image = dia.openImage("DJI000" + str(i) + ".JPG")
print "DJI000" + str(i) + ".JPG"
elif i >= 100 and i < 1000 :
image = dia.openImage("DJI00" + str(i) + ".JPG")
print "DJI00" + str(i) + ".JPG"
else:
image = dia.openImage("DJI0" + str(i) + ".JPG")
print "DJI0" + str(i) + ".JPG"
print "Processing Image" + str(i)
meanVector = dia.meanVector(image)
if meanVector[2] < 80:
print "Correcting Image" + str(i)
newImage = np.array([image[0] + redMeanDifference,image[1],image[2] + meanDifference],dtype='uint8')
x = newImage[0] > image[0]
y = newImage[2] < image[2]
b = newImage[0]
c = newImage[2]
b[x] = 0
c[y] = 255
correctedImage = np.array([b,image[1],c], dtype='uint8')
if i < 10:
dia.outputImage(correctedImage, "corrected/CorDJI0000" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI0000" + str(i) + ".JPG", "DJI0000" + str(i) + ".JPG")
elif i >= 10 and i < 100:
dia.outputImage(correctedImage, "corrected/CorDJI000" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI000" + str(i) + ".JPG", "DJI000" + str(i) + ".JPG")
elif i >= 100 and i < 1000 :
dia.outputImage(correctedImage, "corrected/CorDJI00" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI00" + str(i) + ".JPG", "DJI00" + str(i) + ".JPG")
else:
dia.outputImage(correctedImage, "corrected/CorDJI0" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI0" + str(i) + ".JPG", "DJI0" + str(i) + ".JPG")
else:
if i < 10:
dia.outputImage(image, "corrected/CorDJI0000" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI0000" + str(i) + ".JPG", "DJI0000" + str(i) + ".JPG")
elif i >= 10 and i < 100:
dia.outputImage(image, "corrected/CorDJI000" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI000" + str(i) + ".JPG", "DJI000" + str(i) + ".JPG")
elif i >= 100 and i < 1000 :
dia.outputImage(image, "corrected/CorDJI00" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI00" + str(i) + ".JPG", "DJI00" + str(i) + ".JPG")
else:
dia.outputImage(image, "corrected/CorDJI0" + str(i) + ".JPG")
dia.georefImage("corrected/CorDJI0" + str(i) + ".JPG", "DJI0" + str(i) + ".JPG")
except AttributeError as e:
print e
print "DJI00" + str(i) + ".JPG not found"
