def test_empty_images_histogram_compare_returns_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = None
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
self.assertEqual(img_diff, 'One or more images are empty')
def test_equal_images_histogram_compare_returns_zero(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = im1.copy()
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
self.assertEqual(img_diff, 0)
def parse_disk_image(io_manager: IOManager):
"""
Разбирает образ диска, доступ к которому получен через io_manager
:param io_manager: менеджер, необходимый для работы с образом
:return: FileSystem
"""
info = InfoAboutImage(io_manager)
f_processor = ImageTools.FatProcessor(info, io_manager)
error_detector = ErrorDetector(f_processor)
d_parser = ImageTools.DirectoryParser(f_processor)
ft_printer = ImageTools.FileTreePrinter(d_parser)
if error_detector.check_differences_fats():
return FileSystem(info, f_processor, {}, error_detector)
ft_indexer = ImageTools.FatTableIndexer(d_parser)
full_indexed_fat_table = ft_indexer.get_full_indexed_fat_table()
if error_detector.analysis_fat_indexed_table(full_indexed_fat_table):
return FileSystem(info, f_processor, full_indexed_fat_table,
error_detector)
correct_indexed_fat_table = ft_indexer.get_correct_indexed_fat_table()
file_system = FileSystem(info, f_processor, correct_indexed_fat_table,
error_detector)
file_system.set_file_tree_printer(ft_printer)
return file_system
def test_equal_images_histogram_compare_returns_zero(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = im1.copy()
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
self.assertEqual(img_diff, 0 )
def __init__(self, img_rgb, pattern_name, threshold, probe_func_list):
img_tools = ImageTools()
self.img_rgb = img_rgb
self.threshold = threshold
self.img_gray = img_tools.rgb2gray(self.img_rgb, self.threshold)
self.pattern_name = pattern_name
self.arr_image = img_tools.img2arr(self.img_gray)
self.probe_func_list = probe_func_list
self.lst_near_patterns = [] #...patterns of this class
self.lst_L2 = [] #...distance L2 by each pattern of thi class
self.ave_contrast = 0
self.ave_correlation = 0
self.ave_energy = 0
self.ave_entropy = 0
self.ave_homogeneity = 0
self.l2 = 0
self.class_name = ''
self.class_index = 0
if probe_func_list[0]:
self.contrast = self.calculateContrast(self.arr_image)
if probe_func_list[1]:
self.correlation = self.calculateCorrelation(self.arr_image)
if probe_func_list[2]:
self.energy = self.calculateEnergy(self.arr_image)
if probe_func_list[3]:
self.entropy = self.calculateEntropy(self.arr_image)
if probe_func_list[4]:
self.homogeneity = self.calculateHomogeneity(self.arr_image)
def main():
img_size = 128
classSize = 2000
# Loading model from .pkl
model_file = open('model.pkl', 'rb')
model = pickle.load(model_file)
# Loading data
print("\nImporting data..")
hotdog_files = ImageTools.parseImagePaths('./img/hotdog/')
print("\t..done.\n")
# Preprocess the hotdog files, just like what was done in trainModel.py
# note that the class label isn't necessary, as that is what we're trying to determine.
print("\nGreyscaling and Normalizing Images..")
x, _ = ImageTools.expandClass(hotdog_files, 0, classSize, img_size)
x = np.array(x)
x = ImageTools.greyscaleImgs(x)
x = ImageTools.normalizeImgs(x)
print("\t..done.\n")
# Generating results from the model:
results = model.predict(x)
mean = np.mean(results)
stddev = np.std(results)
print("--")
print("'Is a hotdog a sandwich?''")
print("RESULTS:")
print("\tMean: {}".format(mean))
print("\tStandard Deviation: {}".format(stddev))
print("--")
def test_empty_images_histogram_compare_returns_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = None
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
self.assertEqual(img_diff, 'One or more images are empty' )
def test_opposite_images_pixel_compare_returns_one_hundred(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
img_comp = ImageTools()
img_diff = img_comp._compare_images_pixel(im1, im2)
print(img_diff)
self.assertEqual(img_diff, 100.0)
def test_nonexist_imagefiles_return_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im1.save(self.imfile1)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, 'notreal.jpg')
self.assertEqual(img_diff, 'Error reading one or more files')
def test_opposite_images_histogram_compare_returns_nonzero(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
print(img_diff)
self.assertGreater(img_diff, 0)
def test_opposite_images_histogram_compare_returns_nonzero(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
img_comp = ImageTools()
img_diff = img_comp._compare_images_histogram(im1, im2)
print(img_diff)
self.assertGreater(img_diff, 0)
def write_average_frame(self, filename):
if self.currentAverageFrame is None:
logger.error('Average frame not created')
raise ValueError
#check to see if an error has occured before this
if not self.b_continue:
return
ImageTools.write_image(filename, self.currentAverageFrame)
def test_opposite_images_pixel_compare_returns_one_hundred(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
img_comp = ImageTools()
img_diff = img_comp._compare_images_pixel(im1, im2)
print(img_diff)
self.assertEqual(img_diff, 100.0)
def write_average_frame(self,filename):
if self.currentAverageFrame is None:
logger.error('Average frame not created')
raise ValueError
#check to see if an error has occured before this
if not self.b_continue:
return
ImageTools.write_image(filename, self.currentAverageFrame)
def test_nonexist_imagefiles_return_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im1.save(self.imfile1)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, 'notreal.jpg')
self.assertEqual(img_diff, 'Error reading one or more files')
def test_imagefiles_calls_pixels_algo_by_default(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, self.imfile2)
self.assertEqual(img_diff, 100.0)
def test_imagefiles_calling_histogram_algo(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, self.imfile2, algorithm='histogram')
self.assertGreater(img_diff, 100.0)
def test_imagefiles_calls_pixels_algo_by_default(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, self.imfile2)
self.assertEqual(img_diff, 100.0)
def test_imagefiles_calling_unknown_algo_returns_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1, self.imfile2, algorithm='unknown')
self.assertEqual(img_diff, 'Unsupported algorithm')
def __init__(self, img_rgb, solar_disk_name, threshold, size,
probe_func_list):
img_tools = ImageTools()
self.size = size
self.img_rgb = img_rgb
self.threshold = threshold
self.img_gray = img_tools.rgb2gray(self.img_rgb, self.threshold)
self.pattern_name = solar_disk_name
self.arr_image = img_tools.img2arr(self.img_gray)
self.lst_windows = self.generateFixedWindows(self.arr_image, self.size,
probe_func_list)
def test_imagefiles_calling_unknown_algo_returns_error(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1,
self.imfile2,
algorithm='unknown')
self.assertEqual(img_diff, 'Unsupported algorithm')
def test_imagefiles_calling_histogram_algo(self):
im1 = Image.new('RGB', (256, 256), 'white')
im2 = Image.new('RGB', (256, 256), 'black')
im1.save(self.imfile1)
im2.save(self.imfile2)
img_comp = ImageTools()
img_diff = img_comp.compare_image_files(self.imfile1,
self.imfile2,
algorithm='histogram')
self.assertGreater(img_diff, 100.0)
def computeNextPath(self):
# print('Recomputing path...')
magnitude = 10
# continue straight until we've found a line
if not self.foundLine:
targetPoint = itools.vectorToPoint(self.x, self.y, magnitude, self.degreeRot)
else:
targetPoint = (self.x, self.y)
path = itools.getPixelsBetween(self.x, self.y, targetPoint[0], targetPoint[1])
# print('New path (rot {}) goes from {} to {}'.format(self.degreeRot, (self.x, self.y), (targetPoint[0], targetPoint[1])))
self.stepIter = iter(path) # don't initialize iter to first value
def write_frame(self, filename, frameTypes):
#check to see if an error has occured before this
if self.data is None:
return
if isinstance(frameTypes, str):
frameTypes = [frameTypes]
for frameType in frameTypes:
assert frameType in ['average', 'stdev']
if frameType == 'average':
self.write_average_frame(filename)
if frameType == 'stdev':
if self.currentStdevFrame is not None:
ImageTools.write_image(filename, self.currentStdevFrame)
def write_frame(self,filename,frameTypes):
#check to see if an error has occured before this
if self.data is None:
return
if isinstance(frameTypes,str):
frameTypes = [frameTypes]
for frameType in frameTypes:
assert frameType in ['average','stdev']
if frameType == 'average':
self.write_average_frame(filename)
if frameType == 'stdev':
if self.currentStdevFrame is not None:
ImageTools.write_image(filename,self.currentStdevFrame)
def __init__(self,
path=TIF_IMAGE,
n_samples=N_SAMPLES,
low_res=LOW_RES,
high_res=HIGH_RES):
"""
:param path: the path of the tif file (TODO make it more general)
:param n_samples: the number of wanted samples in the batch.
:param low_res: the low resolution of the 2d image.
:param high_res: the high resolution of the 2d image.
"""
self.path = path
self.im_3d = imread(path)
self.min_d = min(self.im_3d.shape) # the minimal dimension of the 3d
# image
self.n_samples = n_samples
self.low_res = low_res
self.high_res = high_res
self.d_train = self.generate_a_random_batch(1) # test is y slices
self.g_train_hr = self.generate_a_random_batch(0) # train
# is x slices
ImageTools.show_gray_image(self.g_train_hr[0, 0, :, :])
self.g_train_no_cbd = ImageTools.cbd_to_grey(self.g_train_hr)
self.g_train = ImageTools.down_sample(self.g_train_no_cbd)
ImageTools.show_gray_image(self.g_train[0, 0, :, :])
# change both test and train to one hot encoding:
self.ohe_d_train = ImageTools.one_hot_encoding(self.d_train)
self.ohe_g_train = ImageTools.one_hot_encoding(self.g_train)
self.save_batches()
def main():
# 读取配置文件的信息
citycode = Setting.city
if Setting.downwall:
wallpath = '/tmp/auw/bing_wall.jpg'
else:
if len(sys.argv) == 1 or len(sys.argv[1]) == 0:
wallpath = Setting.wallpath
else:
wallpath = sys.argv[1]
config = "# -*-coding:utf-8-*-\n" \
"# 城市代码\n" \
"city = '%s'\n" \
"# 是否下载网络壁纸\n" \
"downwall=%s\n" \
"# 本地壁纸路径\n" \
"wallpath = '%s'" \
% (citycode, 'False', wallpath)
config_file = open('Setting.py', 'w')
config_file.write(config)
config_file.close()
# 设置请求地址与请求参数
request_url = 'http://d1.weather.com.cn/sk_2d/%s.html?_=%s' % (
citycode, int(round(time.time() * 1000)))
request = urllib2.Request(request_url)
request.add_header('referer',
'http://www.weather.com.cn/weather1d/101120101.shtml')
response = urllib2.urlopen(request)
# 读取请求结果并转换为JSON
result = response.read()
result = result.replace('var dataSK = ', '')
result = json.loads(result)
watermark = ImageTools.toWatermarkImage(result)
# 在必应下载壁纸
DownBingWallpaper.downNow(wallpath)
new_wallpath = ImageTools.brand(wallpath, watermark)
commands.getstatusoutput(
"gsettings set org.gnome.desktop.background picture-uri %s" %
new_wallpath)
def make_looped_file(self, name_dir: str):
"""
Создаёт зацикленный файл в директории с названием name_dir, отличной от корневой директории
:param name_dir: название директории, в которой будет создан зацикленный файл
:return: None
"""
empty_entry_point = self._get_free_entry_point_in_dir(name_dir)
free_clusters = ImageTools.find_empty_clusters(
3, self._info, self._file_system.get_indexed_fat_table())
if free_clusters is None:
raise ValueError(
"Not enough free image clusters. Clusters required: " + str(3))
self._dir_parser.create_entry_in_directory(empty_entry_point,
'ERRORLOOP ', 0x00,
free_clusters[0])
for i in range(len(free_clusters)):
if i == len(free_clusters) - 1:
self._ft_proc.write_val_in_all_fat(free_clusters[0],
free_clusters[i])
else:
self._ft_proc.write_val_in_all_fat(free_clusters[i + 1],
free_clusters[i])
def compute_lucky_image(stack):
'''Use the 'lucky' algorithm to compute an average frame from an aligned stack
Original Code Gang Huang, Indiana University, 2010
revised to use convolution by Steve Burns, 2012
Indiana University
this code may be freely distributed and used in publications with attribution to the original paper
reference Huang, G., Zhong, Z. Y., Zou, W. Y., and Burns, S. A., Lucky averaging: quality improvement of adaptive optics scanning laser ophthalmoscope images, Optics Letters 36, 3786-3788 (2011).
'''
nFrames, height, width = stack.shape
numFinal = 15
covStack = np.empty_like(stack) #stack to hold covariance matrices
imageStackSorted = np.empty((numFinal, height, width))
for iFrame in range(nFrames):
covStack[iFrame, :, :] = ImageTools.comatrix(stack[iFrame, :, :])
covStackSortIndex = covStack.argsort(0)
covStackSortIndex[:] = covStackSortIndex[::-1] #reverse the sort
#now resort the image stack so that each pixel in the stack is sorted with the highest contrast pixels in the first frame
for iRow in range(height):
for iCol in range(width):
imageStackSorted[:, iRow,
iCol] = stack[covStackSortIndex[0:numFinal, iRow,
iCol], iRow, iCol]
finalImage = imageStackSorted.mean(axis=0)
return finalImage
def interlaceStack(stack):
"""Attempt to fix poor interlacing in a frame stack
"""
logger.info('Fixing interlace')
nFrames, nRows, nCols = stack.shape
shifts = [ImageTools.getInterlaceShift(stack[iFrame,:,:]) for iFrame in range(nFrames)]
# find the modal shift value
shift = int(max(set(shifts), key=shifts.count))
logger.debug('Shifting interlace by %s pixels',shift)
#allocate a new imageStack
newStack = np.zeros((nFrames,nRows,nCols+abs(shift)))
even_rows = np.arange(0,nRows,2)
odd_rows = np.arange(1,nRows,2)
if shift < 0:
newStack[:,even_rows,abs(shift):newStack.shape[2]] = stack.data[:,even_rows,:]
newStack[:,odd_rows,0:nCols] = stack.data[:,odd_rows,:]
elif shift > 0:
newStack[:,even_rows,0:nCols] = stack.data[:,even_rows,:]
newStack[:,odd_rows,abs(shift):newStack.shape[2]] = stack.data[:,odd_rows,:]
else:
newStack=stack.data
stack.data=newStack
def interlaceStack(stack):
"""Attempt to fix poor interlacing in a frame stack
"""
logger.info('Fixing interlace')
nFrames, nRows, nCols = stack.shape
shifts = [
ImageTools.getInterlaceShift(stack[iFrame, :, :])
for iFrame in range(nFrames)
]
# find the modal shift value
shift = int(max(set(shifts), key=shifts.count))
logger.debug('Shifting interlace by %s pixels', shift)
#allocate a new imageStack
newStack = np.zeros((nFrames, nRows, nCols + abs(shift)))
even_rows = np.arange(0, nRows, 2)
odd_rows = np.arange(1, nRows, 2)
if shift < 0:
newStack[:, even_rows,
abs(shift):newStack.shape[2]] = stack.data[:, even_rows, :]
newStack[:, odd_rows, 0:nCols] = stack.data[:, odd_rows, :]
elif shift > 0:
newStack[:, even_rows, 0:nCols] = stack.data[:, even_rows, :]
newStack[:, odd_rows,
abs(shift):newStack.shape[2]] = stack.data[:, odd_rows, :]
else:
newStack = stack.data
stack.data = newStack
def compute_lucky_image(stack):
'''Use the 'lucky' algorithm to compute an average frame from an aligned stack
Original Code Gang Huang, Indiana University, 2010
revised to use convolution by Steve Burns, 2012
Indiana University
this code may be freely distributed and used in publications with attribution to the original paper
reference Huang, G., Zhong, Z. Y., Zou, W. Y., and Burns, S. A., Lucky averaging: quality improvement of adaptive optics scanning laser ophthalmoscope images, Optics Letters 36, 3786-3788 (2011).
'''
nFrames, height, width = stack.shape
numFinal = 15
covStack = np.empty_like(stack) #stack to hold covariance matrices
imageStackSorted = np.empty((numFinal,height,width))
for iFrame in range(nFrames):
covStack[iFrame,:,:] = ImageTools.comatrix(stack[iFrame,:,:])
covStackSortIndex = covStack.argsort(0)
covStackSortIndex[:] = covStackSortIndex[::-1] #reverse the sort
#now resort the image stack so that each pixel in the stack is sorted with the highest contrast pixels in the first frame
for iRow in range(height):
for iCol in range(width):
imageStackSorted[:,iRow,iCol] = stack[covStackSortIndex[0:numFinal,iRow,iCol],iRow,iCol]
finalImage = imageStackSorted.mean(axis=0)
return finalImage
def fixed_align_frames(self,maxDisplacement=50):
'''perform fixed alignment on the framestack
maxDisplacement=50 - maximum allowed displacement, frames with > than this will be removed from the stack'''
if self.data is None:
logger.warning('No frames found')
return
if self.data.templateFrame is None:
logger.warning('template frame not set')
return
framesToProcess = [i for i in self.data.frameIds if i != self.data.templateFrameId]
midRow = int(self.data.frameWidth / 2)
midCol = int(self.data.frameHeight / 2)
targetFrame = self.data.templateFrame
targetFrame = targetFrame[midRow - self.largeFrameSize : midRow + self.largeFrameSize,
midCol - self.largeFrameSize : midRow + self.largeFrameSize]
results = []
#ensure the target frame is included in the output
results.append({'frameid':self.data.templateFrameId,
'correlation':1,
'shift':(0,0)})
for iFrame in framesToProcess:
templateFrame = self.data.get_frame_by_id(iFrame)
templateFrame = templateFrame[midRow - self.smallFrameSize : midRow + self.smallFrameSize,
midCol - self.smallFrameSize : midCol + self.smallFrameSize]
displacement = ImageTools.find_frame_shift(targetFrame,
templateFrame,
topLeft=[(midRow - self.largeFrameSize,midCol - self.largeFrameSize),
(midRow - self.smallFrameSize,midCol - self.smallFrameSize)],
applyBlur=True,
method='xcorr',
attemptSubPixelAlignment=False)
results.append({'frameid':iFrame,
'correlation':displacement['maxcorr'],
'shift':displacement['coords']})
#Check displacement is les than 50 pixels
good_results = [result for result in results
if abs(result['shift'][1])<=maxDisplacement
and abs(result['shift'][0]) <= maxDisplacement]
bad_results = [result['frameid'] for result in results
if abs(result['shift'][1]) > maxDisplacement
or abs(result['shift'][0]) > maxDisplacement]
logger.info('Removing frames {} for too large displacements'.format(bad_results))
if not good_results:
#no good frames found
logger.warning('frame displacements are too large')
raise RuntimeError('frame displacements are too large')
alignedData = StackTools.apply_displacements(self.data,good_results)
self.data = alignedData
self.currentStack = alignedData
def save_image_url_to_ftp(self, img_url, file_name, session):
ftp_path = 'ImageLib/' + self.manufacturer_long_name + '/'
try:
file = ImageTools.resize_url_for_ebay(img_url)
temp_picture = BytesIO()
file.save(temp_picture, 'jpeg')
temp_picture.seek(0)
session.storbinary("STOR " + ftp_path + file_name, temp_picture)
file.close()
return 'content.powerequipdeals.com/ImageLib/' + self.manufacturer_long_name + '/' + file_name
except:
return ''
def fix_looped_files(self):
"""
Избавляется от найденных зацикленных файлов
:return: None
"""
dir_parser = ImageTools.DirectoryParser(self._fat_proc)
for entry in self.looped_files:
dir_parser.delete_entry_in_directory(
entry.dir_entry_info.entry_point)
self._name_of_indexed_files_to_remove.add(
entry.dir_entry_info.name)
self.looped_files = []
def fix_intersecting_files(self):
"""
Избавляется от найденных пересекающихся файлов
:return: None
"""
dir_parser = ImageTools.DirectoryParser(self._fat_proc)
for list_entries in self.intersecting_files:
for entry in list_entries:
dir_parser.delete_entry_in_directory(
entry.dir_entry_info.entry_point)
self._name_of_indexed_files_to_remove.add(
entry.dir_entry_info.name)
self.intersecting_files = []
def make_intersecting_files(self, name_dir: str):
"""
Создаёт два пересекающихся файла в директории с названием name_dir, отличной от корневой директории
:param name_dir: название директории, в которой будут созданы файлы
:return: None
"""
empty_entry_point = self._get_free_entry_point_in_dir(name_dir)
free_clusters = ImageTools.find_empty_clusters(
3, self._info, self._file_system.get_indexed_fat_table())
if free_clusters is None:
raise ValueError(
"Not enough free image clusters. Clusters required: " + str(3))
self._dir_parser.create_entry_in_directory(empty_entry_point,
'ERRINTERSEC', 0x00,
free_clusters[0])
for i in range(len(free_clusters)):
if i == len(free_clusters) - 1:
self._ft_proc.write_val_in_all_fat(self.end_clus_val,
free_clusters[i])
else:
self._ft_proc.write_val_in_all_fat(free_clusters[i + 1],
free_clusters[i])
empty_entry_point = self._get_free_entry_point_in_dir(name_dir)
new_free_clusters = ImageTools.find_empty_clusters(
1, self._info, self._file_system.get_indexed_fat_table())
if new_free_clusters is None:
raise ValueError(
"Not enough free image clusters. Clusters required: " + str(1))
self._dir_parser.create_entry_in_directory(empty_entry_point,
'ERRINTERS 2', 0x00,
new_free_clusters[0])
self._ft_proc.write_val_in_all_fat(free_clusters[1],
new_free_clusters[0])
def _complete_align_frame(image,frameid):
"""Perform strip alignment on a frame
Expects to be called from complete_align_parallel
Uses some objects placed in shared memory
"""
#start building the output list
result = {'frameid':frameid,'stripResults':[]}
#rebuild the target image from the shared data
targetFrameData = np.asarray(sharedTargetFrameData).reshape(image.shape) #assume that target frame and procFrame are the same shape
smallRowStart = np.asarray(sharedSmallRowStart)
largeRowStart = np.asarray(sharedLargeRowStart)
smallColStart = np.asarray(sharedSmallColStart)[0]
largeColStart = np.asarray(sharedLargeColStart)[0]
largeSzRow = np.asarray(sharedLargeSzRow)[0]
largeSzCol = np.asarray(sharedLargeSzCol)[0]
smallSzRow = np.asarray(sharedSmallSzRow)[0]
smallSzCol = np.asarray(sharedSmallSzCol)[0]
#apply a random mask to the processed frame
mask = np.zeros(image.shape,dtype=np.bool)
mask[image > 0] = 1
image = np.ma.array(image,
mask=~mask)
randomData = image.std() * np.random.standard_normal(image.shape) + image.mean()
image = (image.data * ~image.mask) + (randomData * image.mask) #no longer a masked array
for idxStrip in range(len(sharedSmallRowStart)):
#loop through the strips here
#print('from:{}, to:{}'.format(smallColStart[0],smallColStart[0] + smallSzCol))
smallStrip = image[smallRowStart[idxStrip]:smallRowStart[idxStrip]+smallSzRow,
smallColStart:smallColStart + smallSzCol]
largeStrip = targetFrameData[largeRowStart[idxStrip]:largeRowStart[idxStrip]+largeSzRow,
largeColStart:largeColStart + largeSzCol]
displacement = ImageTools.find_frame_shift(largeStrip,
smallStrip,
topLeft=[(largeRowStart[idxStrip],largeColStart),
(smallRowStart[idxStrip],smallColStart)],
method='xcorr',
applyBlur=True,
attemptSubPixelAlignment=True)
#the offsets returned here are for the small strip within the large strip
#coords = displacement['coords']
#displacement['coords'] = (coords[0] + largeRowStart[idxStrip],
#coords[1] + largeColStart)
result['stripResults'].append(displacement)
logger.debug('done parallel frame{}'.format(frameid))
return result
def __init__(self, nro_line, nro_column, arr_window, probe_func_list):
img_tools = ImageTools()
self.nro_line = nro_line
self.nro_column = nro_column
self.arr_image = arr_window
if probe_func_list[0]:
self.contrast = self.calculateContrast(self.arr_image)
if probe_func_list[1]:
self.correlation = self.calculateCorrelation(self.arr_image)
if probe_func_list[2]:
self.energy = self.calculateEnergy(self.arr_image)
if probe_func_list[3]:
self.entropy = self.calculateEntropy(self.arr_image)
if probe_func_list[4]:
self.homogeneity = self.calculateHomogeneity(self.arr_image)
def getPathsForSpineExtension(self, arcNum, n):
if arcNum >= len(
self.arcPixels): # not initalized with forceInitialized
print("Error: Arc {} hasn't been initialized yet")
return
# get end points of the spine
spineEndPoints = self.getSpineEndPoints(arcNum)
# keep track of the points use for linear regression (for drawing)
allLinRegPoints = set()
# keep track of both paths to return
paths = []
for ep in spineEndPoints: # for each end point
# get which endpoint this is
forwardEnd = len(self.getNextSpinePixels(ep)) == 0
dirKey = "prev" if forwardEnd else "next"
# this endpoints points for linear regression
epLinRegPoints = []
# iterate into the arc and "chop off" the last couple points
currPoint = ep
for _ in range(0, POINTS_TO_CUT):
nextPoints = self._getSpineNeighbors(currPoint, dirKey)
if len(nextPoints) > 1:
print("Erro: There was more than one neighbor.")
currPoint = nextPoints[0]
# currPoint is the new endpoint to be used for linreg
for _ in range(0, POINTS_FOR_LINREG):
epLinRegPoints.append(currPoint)
allLinRegPoints.add(currPoint)
nextPoints = self._getSpineNeighbors(currPoint, dirKey)
if len(nextPoints) > 1:
print("Erro: There was more than one neighbor.")
currPoint = nextPoints[0]
# get n-length path from the endpoint using the linregpoints
epLinRegPoints.reverse() # points added in the wrong direction
path, r2 = itools.interpolateToPath(epLinRegPoints, n, ep)
paths.append(path)
# now we have two n-length paths extending out of both endpoints
return paths
def main():
img_size = 128
classSize = 5000
num_epochs = 15
# Loading Data
print("\nImporting data..")
food_files = ImageTools.parseImagePaths('./img/food/')
sandwich_files = ImageTools.parseImagePaths('./img/sandwich/')
print("\t..done.\n")
print("\nAssigning Labels, Generating more images via transformation..")
print("\tParsing/Labeling foods (sandwiches exclusive)..")
food_x, food_y = ImageTools.expandClass(food_files, 0, classSize, img_size)
print("\t\t..done.")
print("\tParsing/Labeling sandwiches..")
sandwich_x, sandwich_y = ImageTools.expandClass(sandwich_files, 1,
classSize, img_size)
print("\t\t..done.\n")
# Arranging
X = np.array(food_x + sandwich_x)
y = np.array(food_y + sandwich_y)
# Greyscaling and normalizing inputs to reduce features and improve comparability
print("\nGreyscaling and Normalizing Images..")
X = ImageTools.greyscaleImgs(X)
X = ImageTools.normalizeImgs(X)
print("\t..done.\n")
y = to_categorical(y)
# Train n' test:
print("\nSplitting data into training and testing..")
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=np.random.randint(0, 100))
print("\t..done.\n")
print("\tCalling model..")
model = network(img_size) # Calling of CNN
model.compile('adam', 'categorical_crossentropy', ['accuracy'])
print("\t..done.\n")
print("\nTraining Model..")
model.fit(X_train, y_train, nb_epoch=num_epochs, validation_split=0.1)
print("\t..done.\n")
# Saving model
print("\nPickling and saving model as 'model.pkl'...")
modelsave = open('model.pkl', 'wb')
pickle.dump(model, modelsave)
modelsave.close()
print("\t..done.\n")
def compute_variance_image(image,medImage, idx = None):
"""Compute the variance image of a frame
medImage - an ndarray of same size as image with stack median values for each pixel, contains np.nan values outside of the mask region
"""
assert len(image.shape)==2, 'Expected an height x width image'
assert np.all(np.equal(image.shape,medImage.shape))
logger.debug('calulating variance for frame: {}'.format(idx))
image = image.astype(np.float32) #medianBlur function only works on 32bit floats
image = cv2.medianBlur(image,3)
#mask the image
medMask = image > 0
medMask = cv2.erode(medMask.astype(np.float32),np.ones((3,3)),1)
image = image * medMask
image = image + (np.logical_not(medMask) * medImage)
image = ImageTools.unsharp(image,21,3)
if idx is not None:
return [idx, image]
else:
return image
def set_mask(self,roi=None):
"""Create a mask for the image
Params:
roi - [(x1,y1),(x2,y2)]
If roi is None user is prompted to draw a mask, otherwise mask is created from roi"""
if roi is None:
mask = ImageTools.click_and_crop(self.data[0:,:,],types=['mask'])
self.mask = mask['mask']
else:
x1,y1 = roi[0]
x2,y2 = roi[1]
assert x1 >= 0
assert x2 >= x1 and x2 <= self.data.frameWidth
assert y1 >= 0
assert y2 >= y1 and y2 <= self.data.frameHeight
mask = np.zeros((self.data.frameHeight,self.data.frameWidth),dtype=np.bool)
mask[y1:y2, x1:x2] = 1
self.mask = mask
def set_mask(self, roi=None):
"""Create a mask for the image
Params:
roi - [(x1,y1),(x2,y2)]
If roi is None user is prompted to draw a mask, otherwise mask is created from roi"""
if roi is None:
mask = ImageTools.click_and_crop(self.data[0:, :, ],
types=['mask'])
self.mask = mask['mask']
else:
x1, y1 = roi[0]
x2, y2 = roi[1]
assert x1 >= 0
assert x2 >= x1 and x2 <= self.data.frameWidth
assert y1 >= 0
assert y2 >= y1 and y2 <= self.data.frameHeight
mask = np.zeros((self.data.frameHeight, self.data.frameWidth),
dtype=np.bool)
mask[y1:y2, x1:x2] = 1
self.mask = mask
def handleConnection(camera, connection):
# Read the input and send the data
detectColor = (0 ,0, 255)
pixel = ImageTools.detectColor(camera, detectColor)
direction = float(pixel[0]) / camera.resolution[0] * 2 - 1
# Create left speed factor and increase when robot is turning left
leftSpeedFactor = 1
if direction > 0:
leftSpeedFactor = 1 + direction
# Create right speed factor and increase when robot is turning right
rightSpeedFactor = 1
if direction < 0:
rightSpeedFactor = 1 - direction
# Create left and right speeds
leftSpeed = int(leftSpeedFactor * motorSpeed)
rightSpeed = int(rightSpeedFactor * motorSpeed)
print(str(leftSpeed) + ', ' + str(rightSpeed))
# Send the values to the ev3 connection
connection.send('left run ' + str(leftSpeed))
connection.send('right run ' + str(rightSpeed))
def filter_frames(self, minCorr=0.38):
'''Perform an initial filtering on a frame set
filterFrames()
minCorr default 0.38 is ok for cones, other structures require lower values
'''
#check to see if an error has occured before this
if not self.b_continue:
return
framestack = self.data
# calculate mean brightness for each frame,
# if framestack is a masked array masked values are ignored
frame_brightnesses = np.apply_over_axes(np.mean, framestack, [1,2]).flatten()
max_bright = frame_brightnesses.max()
#find frame index for frames >50% max_bright
# frames not in this list will be excluded
good_brights = np.array(frame_brightnesses > max_bright * 0.5, dtype=np.bool)
#good_brights = [i for i, val in enumerate(frame_brightnesses) if val > 0.5* max_bright]
brightestFrames = np.array(frame_brightnesses > max_bright * 0.85, dtype=np.bool)
framelist = np.where(good_brights)[0]
framestack.filter_frames_by_idx(good_brights) # only going to use good frames from here on.
if len(good_brights) < 1:
logger.error("No good frames found")
self.data = None
raise RuntimeError("No good frames found:Brightness too low")
results = []
midRow = int(framestack.frameWidth / 2)
midCol = int(framestack.frameHeight / 2)
for iFrame in np.arange(1,len(framelist)):
currImage = framestack[iFrame - 1,:,:] #target frame
tempImage = framestack[iFrame,:,:] #template frame
shear = ImageTools.find_frame_shear(currImage, tempImage)
tempImage = tempImage[midRow - self.templateSize / 2 : midRow + self.templateSize / 2,
midCol - self.templateSize / 2 : midCol + self.templateSize / 2]
displacement = ImageTools.find_frame_shift(currImage,
tempImage,
topLeft=[(0,0),
(midRow - self.templateSize / 2,midCol - self.templateSize / 2)],
method='xcorr',
applyBlur=True,
attemptSubPixelAlignment=False)
motion = (displacement['coords'][0]**2 + displacement['coords'][0]**2)**0.5
results.append({'frameid':framestack.frameIds[iFrame],
'shear':shear['shearval'],
'correlation':displacement['maxcorr'],
'shift':displacement['coords'],
'motion':motion})
#filter frames where sheer > 20
#results = [r for r in results if r['motion'] <= 20]
#if len(results) < 1:
#logger.error("No good frames found")
#self.data = None
#raise RuntimeError("No good frames found:Shear too high")
#data for frame 0 is missing, use the data from the first remaining frame
#r=[r for r in results if r['frameid'] == 1]
if not results:
raise RuntimeError('Could not get displacements')
r =dict(results[0]) #make a copy of this item
r['frameid']=framelist[0]
r['shift']=(0,0)
r['motion']=0
results.append(r)
maxCorr = max([result['correlation'] for result in results])
if maxCorr < minCorr:
#all correlations are crummy, just bail
#TODO
logger.warning('No good frames found')
raise RuntimeError("No good frames found:Correlation too low")
else:
goodFrames = [result['frameid'] for result in results if result['shear'] < 20 and result['correlation'] > 0.5 * maxCorr and result['motion'] < 50 ]
badFrames = [frameid for frameid in self.data.frameIds if frameid not in goodFrames]
if not goodFrames:
logger.warning('No good frames found')
raise RuntimeError('No good frames found:Group criteria (Shear, Correlation, Motion) not met.')
logger.info('Removing frames {} due to brightness or shear'.format(badFrames))
self.data.filter_frames_by_id(goodFrames)
self.data.templateFrameId = goodFrames[frame_brightnesses[goodFrames].argmax()] #return the brightest of the remaining frames as a potential template
self.filterResults = results #store this for debugging
def complete_align(self,minCorr = 0.38):
"""Takes a roughly aligned stack and performs a complete alignment
minCorr (default 0.38, minimum correlation for inclusion in the output stack)
"""
if self.data is None:
logger.warning('Aborting:No good frames found')
return
nrows,ncols = self.data.frameHeight, self.data.frameWidth
targetFrameData = self.data.templateFrame
framesToProcess = [frameid for frameid in self.data.frameIds if not frameid == self.data.templateFrameId]
#apply a mask to the target frame
mask = np.zeros(targetFrameData.shape,dtype=np.bool)
mask[targetFrameData > 0] = 1
#convert the targetFrameData to a masked array for simple calculation of means
targetFrameData = np.ma.array(targetFrameData,
mask=~mask)
randomData = targetFrameData.std() * np.random.standard_normal(targetFrameData.shape) + targetFrameData.mean()
targetFrameData = (targetFrameData.data * ~targetFrameData.mask) + (randomData * targetFrameData.mask) #no longer a masked array
#setup the row indices
defaultStep = int((nrows - self.smallSzRow + 1) / (self.numberPointsToAlign))
smallRowStart = np.array(range(self.numberPointsToAlign)) * defaultStep
#the large rows should be centered on the small rows
halfDifference = int((self.largeSzRow - self.smallSzRow) / 2)
largeRowStart = smallRowStart - halfDifference # this gives some values out of bounds
largeRowStart[largeRowStart < 0] = 0
maxRowStart = nrows - self.largeSzRow
largeRowStart[largeRowStart > maxRowStart] = maxRowStart
smallColStart = (ncols / 2) - (self.smallSzCol / 2)
largeColStart = (ncols / 2) - (self.largeSzCol / 2)
results = []
for frameId in framesToProcess:
#loop through all the frames here
#need to generate a new mask for each frame
image = self.data.get_frame_by_id(frameId)
mask = np.zeros(image.shape,dtype=np.bool)
mask[image > 0] = 1
image = np.ma.array(image,
mask=~mask)
randomData = image.std() * np.random.standard_normal(image.shape) + image.mean()
image = (image.data * ~image.mask) + (randomData * image.mask) #no longer a masked array
results.append({'frameid':frameId,'stripResults':[]})
for idxStrip in range(len(smallRowStart)):
#loop through the strips here
stripResults = [result['stripResults'] for result in results if result['frameid'] == frameId][0]
smallStrip = image[smallRowStart[idxStrip]:smallRowStart[idxStrip]+self.smallSzRow,
smallColStart:smallColStart + self.smallSzCol]
largeStrip = targetFrameData[largeRowStart[idxStrip]:largeRowStart[idxStrip]+self.largeSzRow,
largeColStart:largeColStart + self.largeSzCol]
displacement = ImageTools.find_frame_shift(largeStrip,
smallStrip,
topLeft=[(largeRowStart[idxStrip],largeColStart),
(smallRowStart[idxStrip],smallColStart)],
method='xcorr',
applyBlur=True,
attemptSubPixelAlignment=True)
#the offsets returned here are for the small strip within the large strip
#coords = displacement['coords']
#displacement['coords'] = (coords[0] + largeRowStart[idxStrip],
#coords[1] + largeColStart)
stripResults.append(displacement)
newCoords = self._get_coords(nrows, ncols)
timetics=[]
for jndx in range(self.numberPointsToAlign):
timetics.append(newCoords['times'][(smallRowStart[jndx]+int(self.smallSzRow/2)),
(smallColStart+int(self.smallSzCol/2)-1)])
self.timeTics = np.array(timetics)
self.times = newCoords['times']
alignmentSplines = self._make_valid_points(results,minCorr)
self.data = self.fast_align(alignmentSplines)
