def processTest(testImageNumList, K, idfList, centers):
"""
This method used to process test images, and transform images to space vector model, and
return testImage list.
"""
testImages = []
processGap = len(testImageNumList)/10.0
processIndex = 0
for testImageIndex in xrange(len(testImageNumList)):
if testImageIndex == len(testImageNumList) - 1:
print ''
elif testImageIndex > processIndex*processGap:
sys.stdout.write('#')
processIndex = processIndex + 1
imageName = readImage.getImageName(testImageNumList[testImageIndex])
SIFTdescr = processSIFT.getSIFT(imageName)
# new train image object, with image name, testImageIndex and SIFT
testImageObject = Image.Image(imageName, testImageIndex, SIFTdescr)
testImages.append(testImageObject)
wordList = []
SIFTdescr = testImages[testImageIndex].SIFTdescr
# get distances between SIFT and centers
distances = getDistances(centers, SIFTdescr)
for siftIndex in xrange(len(testImages[testImageIndex].SIFTdescr)):
# minIndex, minValue = min(enumerate(distances[siftIndex], key = operator.itemgetter(1)))
minIndex = np.argmin(distances[siftIndex])
wordList.append(minIndex)
# save word list into test image
testImages[testImageIndex].setWordList(wordList)
# get tf list
tfList = getTFlist(wordList, K)
svm = getVSM(tfList, idfList)
testImages[testImageIndex].setVSM(svm)
# print svm
return testImages
def processTest(testImageNumList, K, idfList, centers):
"""
This method used to process test images, and transform images to space vector model, and
return testImage list.
"""
testImages = []
processGap = len(testImageNumList) / 10.0
processIndex = 0
for testImageIndex in xrange(len(testImageNumList)):
if testImageIndex == len(testImageNumList) - 1:
print ''
elif testImageIndex > processIndex * processGap:
sys.stdout.write('#')
processIndex = processIndex + 1
imageName = readImage.getImageName(testImageNumList[testImageIndex])
SIFTdescr = processSIFT.getSIFT(imageName)
# new train image object, with image name, testImageIndex and SIFT
testImageObject = Image.Image(imageName, testImageIndex, SIFTdescr)
testImages.append(testImageObject)
wordList = []
SIFTdescr = testImages[testImageIndex].SIFTdescr
# get distances between SIFT and centers
distances = getDistances(centers, SIFTdescr)
for siftIndex in xrange(len(testImages[testImageIndex].SIFTdescr)):
# minIndex, minValue = min(enumerate(distances[siftIndex], key = operator.itemgetter(1)))
minIndex = np.argmin(distances[siftIndex])
wordList.append(minIndex)
# save word list into test image
testImages[testImageIndex].setWordList(wordList)
# get tf list
tfList = getTFlist(wordList, K)
svm = getVSM(tfList, idfList)
testImages[testImageIndex].setVSM(svm)
# print svm
return testImages
def getImages(imageCountList, mleng):
"""
This method used to get the image list which contains basic information,
which contain sift descriptors, image name and edge colorhistogram
[int, int,...], mleng => [image, image,...]
"""
imageList = []
for imageNumber in imageCountList:
imageName = readImage.getImageName(imageNumber)
image = getImageObject(imageName, mleng, imageNumber)
imageList.append(image)
return imageList
def processTrainImages(classSize, K):
"""
This method used to process train images, and return cluster centers, trainImage list and
idfList.
int, int => [[int, int,...],..], trainImages, [int, int,...]
"""
totalDsecr = np.array([])
imageCountList = [1, 2, 3, 4]
trainImages = []
testImageNumList = []
# read train images
for classIndex in xrange(classSize):
randomNum = random.randint(1, 4)
# randomNum = 4
#print 'Find testImage',
#print classIndex*5 + randomNum
testImageNumList.append(classIndex*5 + randomNum)
for imageNumber in imageCountList:
if imageNumber != randomNum:
imageNumber = classIndex*5 + imageNumber
#print 'Find trainImage',
#print imageNumber
imageName = readImage.getImageName(imageNumber)
SIFTdescr = processSIFT.getSIFT(imageName)
# new train image object, with image name, classIndex and SIFT
trainImageObject = Image.Image(imageName, classIndex, SIFTdescr)
trainImages.append(trainImageObject)
# merge all descriptors together
totalDsecr = np.append(totalDsecr, SIFTdescr)
totalDsecr = np.reshape(totalDsecr, (len(totalDsecr)/128, 128))
totalDsecr = np.float32(totalDsecr)
print 'START run Kmeans'
compactness, labels, centers = cv.kmeans(totalDsecr, K, criteria = (cv.TERM_CRITERIA_EPS + cv.TERM_CRITERIA_MAX_ITER, 100000, 0.000001), attempts = 0, flags = cv.KMEANS_RANDOM_CENTERS)
print 'Get compactness is', compactness
print 'FINISH run Kmeans'
start = 0
# now, we assign labels to train images
for trainImageIndex in xrange(len(trainImages)):
descr = trainImages[trainImageIndex].SIFTdescr
trainImages[trainImageIndex].setWordList(labels[start : start + len(descr)])
wordList = trainImages[trainImageIndex].wordList
tfList = getTFlist(wordList, K)
trainImages[trainImageIndex].settfList(tfList)
start = start + len(descr)
# then, we aims to get idf list
idfList = []
for element in xrange(K):
totalValue = 0
for trainImageIndex in xrange(len(trainImages)):
if trainImages[trainImageIndex].tfList[element] > 0:
totalValue = totalValue + 1
idfValue = math.log(len(trainImages)/totalValue)
idfList.append(idfValue)
# then, we build space vector model
for trainImageIndex in xrange(len(trainImages)):
tfList = trainImages[trainImageIndex].tfList
svm = getVSM(tfList, idfList)
trainImages[trainImageIndex].setVSM(svm)
return (centers, trainImages, idfList, testImageNumList)
def processTrainImages(classSize, K):
"""
This method used to process train images, and return cluster centers, trainImage list and
idfList.
int, int => [[int, int,...],..], trainImages, [int, int,...]
"""
totalDsecr = np.array([])
imageCountList = [1, 2, 3, 4]
trainImages = []
testImageNumList = []
# read train images
for classIndex in xrange(classSize):
randomNum = random.randint(1, 4)
# randomNum = 4
#print 'Find testImage',
#print classIndex*5 + randomNum
testImageNumList.append(classIndex * 5 + randomNum)
for imageNumber in imageCountList:
if imageNumber != randomNum:
imageNumber = classIndex * 5 + imageNumber
#print 'Find trainImage',
#print imageNumber
imageName = readImage.getImageName(imageNumber)
SIFTdescr = processSIFT.getSIFT(imageName)
# new train image object, with image name, classIndex and SIFT
trainImageObject = Image.Image(imageName, classIndex,
SIFTdescr)
trainImages.append(trainImageObject)
# merge all descriptors together
totalDsecr = np.append(totalDsecr, SIFTdescr)
totalDsecr = np.reshape(totalDsecr, (len(totalDsecr) / 128, 128))
totalDsecr = np.float32(totalDsecr)
print 'START run Kmeans'
compactness, labels, centers = cv.kmeans(
totalDsecr,
K,
criteria=(cv.TERM_CRITERIA_EPS + cv.TERM_CRITERIA_MAX_ITER, 100000,
0.000001),
attempts=0,
flags=cv.KMEANS_RANDOM_CENTERS)
print 'Get compactness is', compactness
print 'FINISH run Kmeans'
start = 0
# now, we assign labels to train images
for trainImageIndex in xrange(len(trainImages)):
descr = trainImages[trainImageIndex].SIFTdescr
trainImages[trainImageIndex].setWordList(labels[start:start +
len(descr)])
wordList = trainImages[trainImageIndex].wordList
tfList = getTFlist(wordList, K)
trainImages[trainImageIndex].settfList(tfList)
start = start + len(descr)
# then, we aims to get idf list
idfList = []
for element in xrange(K):
totalValue = 0
for trainImageIndex in xrange(len(trainImages)):
if trainImages[trainImageIndex].tfList[element] > 0:
totalValue = totalValue + 1
idfValue = math.log(len(trainImages) / totalValue)
idfList.append(idfValue)
# then, we build space vector model
for trainImageIndex in xrange(len(trainImages)):
tfList = trainImages[trainImageIndex].tfList
svm = getVSM(tfList, idfList)
trainImages[trainImageIndex].setVSM(svm)
return (centers, trainImages, idfList, testImageNumList)