def __init__( self, image_path, env ):
state( 'Opening image...' )
self.env = env
if( self.setImage( image_path ) ):
comment( '- Image opened successfully.')
else:
error( 'Could not set image from path in ProcessImage.' )
def getFeatureFiles(self):
comment('Reading files.')
tmp_list = []
for dir in self._directory_list:
tmp_json = HandleJSON(dir)
tmp_list.append({
"feature_json": tmp_json.file,
"directory": dir,
"filepath": tmp_json.filepath
})
return tmp_list
def setPaths(self, subject: str):
try:
comment('Setting paths for image directories.')
for i in range(len(self._directory_list)):
self._directory_list[
i] = f'{ self._directory_list[ i ] }/{ subject }'
return TestMessage(
True, 'Complete list of paths created for feature extraction.')
except IOError as ioe:
return TestMessage(False, ioe)
except:
return TestMessage(False,
'Could not add subject type to path variables')
def test_comment(self):
"""
Test whether a successful comment can be made to the console.
"""
self.assertEqual(
comment('Testing a comment message').success, True,
'A successful comment message should return a success status of True.'
)
def buff( self, small_image ):
comment( '- Buffing image.' )
big_image = Image.fromarray(
np.uint8(
np.full([
self.env.settings.content[ 'extract' ][ 'height'],
self.env.settings.content[ 'extract' ][ 'width']
],
255)
)
)
small_dim = np.shape( np.array( small_image ) )
big_dim = np.shape( np.array( big_image ) )
coordinates = []
# werk eers op die eerste as.
s_ax = small_dim[ 0 ]
b_ax = big_dim[ 0 ]
if( s_ax == b_ax ):
coordinates.append( 0 )
pass
elif( s_ax > b_ax):
coordinates.append( 0 )
else:
coordinates.append( int( ( b_ax/2 ) - ( s_ax/2 ) ) )
# werk dan op die tweede as
s_ax = small_dim[ 1 ]
b_ax = big_dim[ 1 ]
if( s_ax == b_ax ):
coordinates.append( 0 )
pass
elif( s_ax > b_ax):
coordinates.append( 0 )
else:
coordinates.append( int( ( b_ax/2 ) - ( s_ax/2 ) ) )
big_image.paste( Image.fromarray( small_image ), ( coordinates[ 1 ], coordinates[ 0 ]) )
big_image = np.asarray( big_image )
return big_image
def writeFiles( self, bundle ):
try:
os.makedirs( bundle.getPath() )
os.chdir( bundle.getPath() )
files = Files( bundle.getPath() , self.env )
comment( '- Writing content for subdirectories.' )
files.writeWords( 'words', bundle.words, bundle.folder_name )
files.writeLines( 'lines', bundle.lines, bundle.folder_name )
except:
error( 'Directory already exists.' )
rm_file = input( 'Do you want to remove the directory [Y/n]?' )
if( rm_file.upper() == 'Y' or rm_file == '' ):
os.system( f'rm { bundle.getPath() } -r')
os.makedirs( bundle.getPath() )
os.chdir( bundle.getPath() )
files = Files( bundle.getPath(), self.env )
comment( '- Writing content for subdirectories.' )
files.writeWords( 'words', bundle.words, bundle.folder_name )
files.writeLines( 'lines', bundle.lines, bundle.folder_name )
def getDirectories(self, path):
comment('Reading directories in dataset.')
try:
samples = []
for dir in os.scandir(path):
if (os.DirEntry.is_dir(dir)):
samples.append(dir.path)
if (samples != []):
self._directory_list = samples
return TestMessage(
True,
'Successfully compiled a list of directories for the sample.',
7)
else:
self._directory_list = samples
return TestMessage(
False,
'There was no relevant sample directories in this path.',
7)
except Exception as e:
return TestMessage(False, f'{e}', 22)
def setImage( self, path ):
try:
comment( 'Reading image.' )
comment( '- Loading from path: {0}'.format( path ) )
self.image = cv2.imread( path )
comment( '- Flattening image into matrix.' )
# Flattening and reshaping the array
self.matrix = np.array( cv2.cvtColor( self.image, cv2.COLOR_RGB2GRAY ))
x, y = np.shape( self.matrix )
self.dimensions = [ x, y ]
comment( '- Matrix shape: {0}'.format( str( self.dimensions )))
return True
except:
return False
def __init__(self, env, feature: str, path=None):
state('Setting up the image pipeline.')
self._e = env
if (path != None):
directories = self.getDirectories(path)
if (directories.success):
# Inside each sample folder there are images for lines and words, which are folders.
self.subject = ask('Would you like to use [words] or [lines]?',
readInput.readline())
while (not (self.subject.payload == 'words'
or self.subject.payload == 'lines')):
self.subject = ask('Type either [words] or [lines]?',
readInput.readline())
tmp_create_paths = self.setPaths(self.subject.payload)
if (tmp_create_paths.success):
comment(tmp_create_paths.payload)
feature_files = self.getFeatureFiles()
available_features = self.getListOfAvailableFeatures()
comment('Adding features.')
for feature_class in available_features:
self.save(
self.generateFeatures(feature_files,
feature_class),
feature_class)
else:
error(tmp_create_paths.payload)
raise IOError(tmp_create_paths.payload)
else:
error(directories.payload)
def __init__(self, path: str):
comment(f"Handling JSON entry for {path}.")
self.getJson( path )
def getLines( self ):
print( )
extract_buffer = self.env.settings.content[ 'extract' ][ 'buffer']
kernel_size = self.env.settings.content[ 'extract' ][ 'kernel_size']
comment( 'Extracting text lines.' )
threshold = self.env.settings.content[ 'extract' ][ 'threshold']
# This is the structuring element that will be used in the dilation of the mask.
comment( '- Creating structuring element with a kernel size of {0}.'.format( kernel_size ) )
kernel = cv2.getStructuringElement( cv2.MORPH_DILATE, ( kernel_size, kernel_size ) )
# Eroding image
self.image = cv2.cvtColor( self.image, cv2.COLOR_BGR2GRAY )
ret, mask = cv2.threshold( self.image, threshold, 255, cv2.THRESH_BINARY )
comment( '- Dilating image.' )
mask = cv2.bitwise_not( mask )
original_mask = mask
mask = cv2.dilate( mask, kernel, iterations= 1 )
# Setting up run-length array
comment( '- Extracting text lines.' )
self.matrix = np.array( mask )
regions = []
totals = []
region_flag = False
# Extracting line regions by using horizontal projections
for x in range( 0, self.dimensions[ 0 ] ):
total = 0
for y in range( 0, self.dimensions[ 1 ]):
total += self.matrix[ x ][ y ]/255
totals.append( total )
if( total > extract_buffer and region_flag != True):
begin_region = x
region_flag = True
elif( total <= self.env.settings.content[ 'extract' ][ 'buffer'] and region_flag == True ):
"""
- If you reach the end of the one region the boundaries and sub-image is stored in the regions array to be returned by the function.
- The extra if-statement is to ensure that only relevant segments are stored.
"""
[ seg_width, seg_height ] = np.shape( self.image[ begin_region:x, 0:self.dimensions[ 1 ] ] )
if( seg_width > self.env.settings.content[ 'extract' ][ 'tolerance' ] ):
# resizing the image height to be consistent for the neural network
dim = ( seg_height, self.env.settings.content[ 'extract' ][ 'height' ] )
resized_image = cv2.resize(
self.image[ begin_region:x, 0:self.dimensions[ 1 ] ] ,
dim,
interpolation = cv2.INTER_AREA
)
resized_mask = cv2.resize(
original_mask[ begin_region:x, 0:self.dimensions[ 1 ] ] ,
dim,
interpolation = cv2.INTER_AREA
)
regions.append({
'begin': begin_region,
'end': x,
'image': resized_image,
'mask': resized_mask
})
region_flag = False
return regions
def getWords( self, line ):
( rows, cols ) = np.shape( line[ 'image' ] )
# Erode the mask again to get more gaps in the text.
extract_buffer = self.env.settings.content[ 'extract' ][ 'buffer']
comment( '- Extracting words from line.' )
mask = np.array( line[ 'mask' ] )
words = []
word_flag = False
for c in range( 0, cols ):
total = 0
for r in range( 0, rows ):
total += mask[ r ][ c ]/255
if( total > extract_buffer and word_flag == False ):
word_flag = True
begin_word = c
elif( total == 0 and word_flag == True ):
word_flag = False
# Divide the word image into smaller segments if possible
word = line[ 'image' ][ 0:rows, begin_word:c ]
word_mask = mask[ 0:rows, begin_word:c ]
( height, width ) = np.shape( word )
word_length = c - begin_word
if( word_length + ( self.env.settings.content[ 'extract' ][ 'buffer'] ) <= self.env.settings.content[ 'extract' ][ 'width'] ):
resized_word_image = cv2.resize(
word,
( height, self.env.settings.content[ 'extract' ][ 'width'] ),
interpolation = cv2.INTER_AREA
)
resized_word_mask = cv2.resize(
word_mask,
( height, self.env.settings.content[ 'extract' ][ 'width'] ),
interpolation = cv2.INTER_AREA
)
words.append({
'begin': begin_word,
'end': c,
'image': resized_word_image,
'mask': resized_word_mask
})
else:
div = ceil( width / self.env.settings.content[ 'extract' ][ 'width'] )
count = 0
while( count < width - ( width / div ) ):
seg = word[ 0:rows, count: count + int( width / div ) ]
seg_mask = mask[ 0:rows, count: count + int( width / div ) ]
if( np.shape( seg )[1] == 0): print( count, width )
words.append({
'begin': count,
'end': count + int( width / div ),
'image': seg,
'mask': seg_mask
})
count += int( width / div )
# If there are still some pixels left for the whole word to be extracted.
seg = word[ 0:rows, count: width ]
seg_mask = mask[ 0:rows, count: width ]
if( np.shape( seg )[1] > 0 ):
words.append({
'begin': count,
'end': width,
'image': seg,
'mask': seg_mask
})
return words
def save(self, file, feature_class):
outfile = open(file['filepath'], 'w')
outfile.write(json.dumps(file['feature_json'], indent=4))
outfile.close()
comment(f'Successfully added {feature_class} feature.')
