class JpegEncoder(object):
def __init__(self, image, quality, out, comment, ais):
self.quality = quality
self.jpeg_obj = JpegInfo(image, comment)
self.image_width, self.image_height = image.size
self.out = out
self.dct = DCT(self.quality)
self.huf = Huffman(*image.size)
self.hasais = ais
self.k_matrix = -1
def compress(self, embedded_data=None, password='******'):
self.embedded_data = EmbedData(
embedded_data) if embedded_data else None
self.password = password
self.write_headers()
self.write_compressed_data()
self.write_eoi()
self.out.flush()
def get_quality(self):
return self.quality
def set_quality(self, quality):
self.quality = quality
self.dct = DCT(quality)
def write_array(self, data):
length = ((data[2] & 0xff) << 8) + (data[3] & 0xff) + 2
self.out.write(bytearray(data[:length]))
def write_marker(self, data):
self.out.write(bytearray(data[:2]))
def write_eoi(self):
EOI = [0xff, 0xD9]
self.write_marker(EOI)
def write_headers(self):
SOI = [0xff, 0xD8]
self.write_marker(SOI)
JFIF = [
0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01, 0x01,
0x01, 0x00, 0x60, 0x00, 0x60, 0x00, 0x00
]
self.write_array(JFIF)
comment = self.jpeg_obj.get_comment()
if comment:
length = len(comment) + 2
COM = [0xff, 0xfe, length >> 8 & 0xff, length & 0xff]
COM.extend(comment)
self.write_array(COM)
DQT = [0xff, 0xdb, 0x00, 0x84]
for k in range(2):
DQT.append(k)
DQT.extend([
self.dct.quantum[k][JPEG_NATURAL_ORDER[i]] for i in range(64)
])
self.write_array(DQT)
SOF = [
0xff, 0xc0, 0x00, 0x11, self.jpeg_obj.precision,
self.jpeg_obj.image_height >> 8 & 0xff,
self.jpeg_obj.image_height & 0xff,
self.jpeg_obj.image_width >> 8 & 0xff,
self.jpeg_obj.image_width & 0xff, self.jpeg_obj.comp_num
]
for i in range(self.jpeg_obj.comp_num):
SOF.append(self.jpeg_obj.com_id[i])
SOF.append(
eight_byte(self.jpeg_obj.hsamp_factor[i],
self.jpeg_obj.vsamp_factor[i]))
SOF.append(self.jpeg_obj.qtable_number[i])
self.write_array(SOF)
DHT = [0xff, 0xc4, 0, 0]
for i in range(4):
DHT.extend(self.huf.BITS[i])
DHT.extend(self.huf.VAL[i])
DHT[2] = len(DHT) - 2 >> 8 & 0xff
DHT[3] = len(DHT) - 2 & 0xff
self.write_array(DHT)
SOS = [0] * 14
SOS = [0xff, 0xda, 0x00, 0x0c, self.jpeg_obj.comp_num]
for i in range(self.jpeg_obj.comp_num):
SOS.append(self.jpeg_obj.com_id[i])
SOS.append(
eight_byte(self.jpeg_obj.dctable_number[i],
self.jpeg_obj.actable_number[i]))
SOS.append(self.jpeg_obj.ss)
SOS.append(self.jpeg_obj.se)
SOS.append(eight_byte(self.jpeg_obj.ah, self.jpeg_obj.al))
self.write_array(SOS)
def _get_coeff(self):
dct_array1 = create_array(0.0, 8, 8)
dct_array2 = create_array(0.0, 8, 8)
dct_array3 = create_array(0, 64)
coeff = []
for r in range(min(self.jpeg_obj.block_height)):
for c in range(min(self.jpeg_obj.block_width)):
xpos = c * 8
ypos = r * 8
for comp in range(self.jpeg_obj.comp_num):
indata = self.jpeg_obj.components[comp]
maxa = self.image_height / 2 * self.jpeg_obj.vsamp_factor[
comp] - 1
maxb = self.image_width / 2 * self.jpeg_obj.hsamp_factor[
comp] - 1
for i in range(self.jpeg_obj.vsamp_factor[comp]):
for j in range(self.jpeg_obj.hsamp_factor[comp]):
ia = ypos * self.jpeg_obj.vsamp_factor[comp] + i * 8
ib = xpos * self.jpeg_obj.hsamp_factor[comp] + j * 8
for a in range(8):
for b in range(8):
dct_array1[a][b] = indata[min(
ia + a, maxa)][min(ib + b, maxb)]
dct_array2 = self.dct.forward_dct(dct_array1)
dct_array3 = self.dct.quantize_block(
dct_array2, self.jpeg_obj.qtable_number[comp])
coeff.extend(dct_array3[:64])
return coeff
def write_compressed_data(self):
last_dc_value = create_array(0, self.jpeg_obj.comp_num)
zero_array = create_array(0, 64)
width, height = 0, 0
min_block_width = min(self.jpeg_obj.block_width)
min_block_height = min(self.jpeg_obj.block_height)
logger.info('DCT/quantisation starts')
logger.info('%d x %d' % (self.image_width, self.image_height))
coeff = self._get_coeff() #量化后的系数,是整数
coeff_count = len(coeff)
#导出未处理的QDCT
if self.hasais:
filename = 'unpro_ais.json'
else:
filename = 'unpro.json'
with open(filename, 'w') as f_unprocess:
json.dump(coeff, f_unprocess)
#AIS处理
if self.hasais:
size_secret = self.embedded_data.len
ais = Ais(coeff, size_secret) #coeff被修改
ais.statistic()
self.k_matrix = ais.fix()
with open('aised.json', 'w') as f_aised:
json.dump(coeff, f_aised)
#嵌入——>再统计嵌入后的数据,决定是否继续做AIS处理
logger.info('got %d DCT AC/DC coefficients' % coeff_count)
_changed, _embedded, _examined, _expected, _one, _large, _thrown, _zero = 0, 0, 0, 0, 0, 0, 0, 0
shuffled_index = 0
for i, cc in enumerate(coeff):
if i % 64 == 0:
continue
if cc == 1 or cc == -1:
_one += 1
elif cc == 0:
_zero += 1
_large = coeff_count - _zero - _one - coeff_count / 64 #有效系数的个数
_expected = _large + int(0.49 * _one) #预期容量,shrinkage效应无法确定
logger.info('one=%d' % _one)
logger.info('large=%d' % _large)
logger.info('\nexpected capacity: %d bits\n' % _expected)
logger.info('expected capacity with')
for i in range(1, 8):
n = (1 << i) - 1 #n=2^i-1
changed = _large - _large % (n + 1)
changed = (changed + _one + _one / 2 - _one / (n + 1)) / (n + 1)
usable = (_expected * i / n - _expected * i / n % n) / 8
if usable == 0:
break
logger.info(
'%s code: %d bytes (efficiency: %d.%d bits per change)' %
('default' if i == 1 else '(1, %d, %d)' % (n, i), usable,
usable * 8 / changed, usable * 80 / changed % 10))
#shuffles all coefficients using a permutation,使用排列对系数进行混洗
if self.embedded_data is not None:
logger.info('permutation starts')
random = F5Random(self.password)
permutation = Permutation(coeff_count, random)
next_bit_to_embed = 0
byte_to_embed = len(self.embedded_data)
available_bits_to_embed = 0
logger.info('Embedding of %d bits (%d+4 bytes)' %
(byte_to_embed * 8 + 32, byte_to_embed))
if byte_to_embed > 0x007fffff:
byte_to_embed = 0x007ffff
for i in range(1, 8):
self.n = (1 << i) - 1
usable = (_expected * i / self.n -
_expected * i / self.n % self.n) / 8
if usable < byte_to_embed + 4:
break
#确定(1,n,k)
if self.k_matrix < 0:
k = i - 1
else:
k = self.k_matrix
self.n = (1 << k) - 1
if self.n == 0:
logger.info('using default code, file will not fit')
self.n = 1
elif self.n == 1:
logger.info('using default code')
else:
logger.info('using (1, %d, %d) code' % (self.n, k))
byte_to_embed |= k << 24
byte_to_embed ^= random.get_next_byte()
byte_to_embed ^= random.get_next_byte() << 8
byte_to_embed ^= random.get_next_byte() << 16
byte_to_embed ^= random.get_next_byte() << 24
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed = 31
_embedded += 1
for i, shuffled_index in enumerate(permutation.shuffled):
if shuffled_index % 64 == 0 or coeff[shuffled_index] == 0:
continue
cc = coeff[shuffled_index]
_examined += 1
if cc > 0 and (cc & 1) != next_bit_to_embed:
coeff[shuffled_index] -= 1
_changed += 1
elif cc < 0 and (cc & 1) == next_bit_to_embed:
coeff[shuffled_index] += 1
_changed += 1
if coeff[shuffled_index] != 0:
if available_bits_to_embed == 0:
if self.n > 1 or not self.embedded_data.available():
break
byte_to_embed = self.embedded_data.read()
byte_to_embed ^= random.get_next_byte()
available_bits_to_embed = 8
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed -= 1
_embedded += 1
else:
_thrown += 1
if self.n > 1:
try:
is_last_byte = False
filtered_index = FilteredCollection(
permutation.shuffled[i + 1:],
lambda index: index % 64 and coeff[index])
while not is_last_byte:
k_bits_to_embed = 0
for i in range(k):
if available_bits_to_embed == 0:
if not self.embedded_data.available():
is_last_byte = True
break
byte_to_embed = self.embedded_data.read()
byte_to_embed ^= random.get_next_byte()
available_bits_to_embed = 8
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed -= 1
k_bits_to_embed |= next_bit_to_embed << i
_embedded += 1
code_word = filtered_index.offer(self.n)
while True:
vhash = 0
for i, index in enumerate(code_word):
if coeff[index] > 0:
extracted_bit = coeff[index] & 1
else:
extracted_bit = 1 - (coeff[index] & 1)
if extracted_bit == 1:
vhash ^= i + 1
i = vhash ^ k_bits_to_embed
if not i:
break
i -= 1
coeff[code_word[i]] += 1 if coeff[
code_word[i]] < 0 else -1
_changed += 1
if not coeff[code_word[i]]:
_thrown += 1
code_word[i:i + 1] = []
code_word.extend(filtered_index.offer(1))
else:
break
except FilteredCollection.ListNotEnough:
pass
if _examined > 0:
logger.info('%d coefficients examined' % _examined)
if _changed > 0:
logger.info(
'%d coefficients changed (efficiency: %d.%d bits per change'
% (_changed, _embedded / _changed,
_embedded * 10 / _changed % 10))
logger.info('%d coefficients thrown (zeroed)' % _thrown)
logger.info('%d bits (%d bytes) embedded' %
(_embedded, _embedded / 8))
#导出嵌入后的系数coeff
if self.hasais:
filename2 = 'embeded_ais.json'
else:
filename2 = 'embeded.json'
with open(filename2, 'w') as f_embeded:
json.dump(coeff, f_embeded)
logger.info('starting hufman encoding')
shuffled_index = 0
for r in range(min_block_height):
for c in range(min_block_width):
for comp in range(self.jpeg_obj.comp_num):
for i in range(self.jpeg_obj.vsamp_factor[comp]):
for j in range(self.jpeg_obj.hsamp_factor[comp]):
dct_array3 = coeff[shuffled_index:shuffled_index +
64]
self.huf.huffman_block_encoder(
self.out, dct_array3, last_dc_value[comp],
self.jpeg_obj.dctable_number[comp],
self.jpeg_obj.actable_number[comp])
last_dc_value[comp] = dct_array3[0]
shuffled_index += 64
self.huf.flush_buffer(self.out)
logger.info('hufman encode end')
class JpegEncoder(object):
def __init__(self, image, quality, out, comment):
self.quality = quality
self.jpeg_obj = JpegInfo(image, comment)
self.image_width, self.image_height = image.size
self.out = out
self.dct = DCT(self.quality)
self.huf = Huffman(*image.size)
def compress(self, embedded_data=None, password='******'):
self.embedded_data = EmbedData(embedded_data) if embedded_data else None
self.password = password
self.write_headers()
self.write_compressed_data()
self.write_eoi()
self.out.flush()
def get_quality(self):
return self.quality
def set_quality(self, quality):
self.quality = quality
self.dct = DCT(quality)
def write_array(self, data):
length = ((data[2] & 0xff) << 8) + (data[3] & 0xff) + 2
self.out.write(bytearray(data[:length]))
def write_marker(self, data):
self.out.write(bytearray(data[:2]))
def write_eoi(self):
EOI = [0xff, 0xD9]
self.write_marker(EOI)
def write_headers(self):
SOI = [0xff, 0xD8]
self.write_marker(SOI)
JFIF = [0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46,
0x49, 0x46, 0x00, 0x01, 0x01, 0x01,
0x00, 0x60, 0x00, 0x60, 0x00, 0x00]
self.write_array(JFIF)
comment = self.jpeg_obj.get_comment()
if comment:
length = len(comment) + 2
COM = [0xff, 0xfe, length >> 8 & 0xff, length & 0xff]
COM.extend(comment)
self.write_array(COM)
DQT = [0xff, 0xdb, 0x00, 0x84]
for k in range(2):
DQT.append(k)
DQT.extend([self.dct.quantum[k][JPEG_NATURAL_ORDER[i]] for i in range(64)])
self.write_array(DQT)
SOF = [0xff, 0xc0, 0x00, 0x11,
self.jpeg_obj.precision,
self.jpeg_obj.image_height >> 8 & 0xff,
self.jpeg_obj.image_height & 0xff,
self.jpeg_obj.image_width >> 8 & 0xff,
self.jpeg_obj.image_width & 0xff,
self.jpeg_obj.comp_num]
for i in range(self.jpeg_obj.comp_num):
SOF.append(self.jpeg_obj.com_id[i])
SOF.append(eight_byte(self.jpeg_obj.hsamp_factor[i], self.jpeg_obj.vsamp_factor[i]))
SOF.append(self.jpeg_obj.qtable_number[i])
self.write_array(SOF)
DHT = [0xff, 0xc4, 0, 0]
for i in range(4):
DHT.extend(self.huf.BITS[i])
DHT.extend(self.huf.VAL[i])
DHT[2] = len(DHT) - 2 >> 8 & 0xff
DHT[3] = len(DHT) - 2 & 0xff
self.write_array(DHT)
SOS = [0] * 14
SOS = [0xff, 0xda, 0x00, 0x0c, self.jpeg_obj.comp_num]
for i in range(self.jpeg_obj.comp_num):
SOS.append(self.jpeg_obj.com_id[i])
SOS.append(eight_byte(self.jpeg_obj.dctable_number[i], self.jpeg_obj.actable_number[i]))
SOS.append(self.jpeg_obj.ss)
SOS.append(self.jpeg_obj.se)
SOS.append(eight_byte(self.jpeg_obj.ah, self.jpeg_obj.al))
self.write_array(SOS)
def _get_coeff(self):
dct_array1 = create_array(0.0, 8, 8)
dct_array2 = create_array(0.0, 8, 8)
dct_array3 = create_array(0, 64)
coeff = []
for r in range(min(self.jpeg_obj.block_height)):
for c in range(min(self.jpeg_obj.block_width)):
xpos = c * 8
ypos = r * 8
for comp in range(self.jpeg_obj.comp_num):
indata = self.jpeg_obj.components[comp]
maxa = self.image_height / 2 * self.jpeg_obj.vsamp_factor[comp] - 1
maxb = self.image_width / 2 * self.jpeg_obj.hsamp_factor[comp] - 1
for i in range(self.jpeg_obj.vsamp_factor[comp]):
for j in range(self.jpeg_obj.hsamp_factor[comp]):
ia = ypos * self.jpeg_obj.vsamp_factor[comp] + i * 8
ib = xpos * self.jpeg_obj.hsamp_factor[comp] + j * 8
for a in range(8):
for b in range(8):
dct_array1[a][b] = indata[min(ia+a, maxa)][min(ib+b, maxb)]
dct_array2 = self.dct.forward_dct(dct_array1)
dct_array3 = self.dct.quantize_block(dct_array2,
self.jpeg_obj.qtable_number[comp])
coeff.extend(dct_array3[:64])
return coeff
def write_compressed_data(self):
tmp = 0
last_dc_value = create_array(0, self.jpeg_obj.comp_num)
zero_array = create_array(0, 64)
width, height = 0, 0
min_block_width = min(self.jpeg_obj.block_width)
min_block_height = min(self.jpeg_obj.block_height)
logger.info('DCT/quantisation starts')
logger.info('%d x %d' % (self.image_width, self.image_height))
coeff = self._get_coeff()
coeff_count = len(coeff)
logger.info('got %d DCT AC/DC coefficients' % coeff_count)
_changed, _embedded, _examined, _expected, _one, _large, _thrown, _zero = 0, 0, 0, 0, 0, 0, 0, 0
shuffled_index = 0
for i, cc in enumerate(coeff):
if i % 64 == 0:
continue
if cc == 1 or cc == -1:
_one += 1
elif cc == 0:
_zero += 1
_large = coeff_count - _zero - _one - coeff_count / 64
_expected = _large + int(0.49 * _one)
logger.info('one=%d' % _one)
logger.info('large=%d' % _large)
logger.info('expected capacity: %d bits' % _expected)
logger.info('expected capacity with')
for i in range(1, 8):
n = (1 << i) - 1
changed = _large - _large % (n + 1)
changed = (changed + _one + _one / 2 - _one / (n + 1)) / (n + 1)
usable = (_expected * i / n - _expected * i / n % n) / 8
if usable == 0:
break
logger.info('%s code: %d bytes (efficiency: %d.%d bits per change)' % ('default' if i == 1 else '(1, %d, %d)' % (n, i), usable, usable * 8 / changed, usable * 80 / changed % 10))
if self.embedded_data is not None:
logger.info('permutation starts')
random = F5Random(self.password)
permutation = Permutation(coeff_count, random)
next_bit_to_embed = 0
byte_to_embed = len(self.embedded_data)
available_bits_to_embed = 0
logger.info('Embedding of %d bits (%d+4 bytes)' % (byte_to_embed * 8 + 32, byte_to_embed))
if byte_to_embed > 0x007fffff:
byte_to_embed = 0x007ffff
for i in range(1, 8):
self.n = (1 << i) - 1
usable = (_expected * i / self.n - _expected * i / self.n % self.n) / 8
if usable < byte_to_embed + 4:
break
k = i - 1
self.n = (1 << k) - 1
if self.n == 0:
logger.info('using default code, file will not fit')
self.n = 1
elif self.n == 1:
logger.info('using default code')
else:
logger.info('using (1, %d, %d) code' % (self.n, k))
byte_to_embed |= k << 24
byte_to_embed ^= random.get_next_byte()
byte_to_embed ^= random.get_next_byte() << 8
byte_to_embed ^= random.get_next_byte() << 16
byte_to_embed ^= random.get_next_byte() << 24
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed = 31
_embedded += 1
for i, shuffled_index in enumerate(permutation.shuffled):
if shuffled_index % 64 == 0 or coeff[shuffled_index] == 0:
continue
cc = coeff[shuffled_index]
_examined += 1
if cc > 0 and (cc & 1) != next_bit_to_embed:
coeff[shuffled_index] -= 1
_changed +=1
elif cc < 0 and (cc & 1) == next_bit_to_embed:
coeff[shuffled_index] += 1
_changed += 1
if coeff[shuffled_index] != 0:
if available_bits_to_embed == 0:
if self.n > 1 or not self.embedded_data.available():
break
byte_to_embed = self.embedded_data.read()
byte_to_embed ^= random.get_next_byte()
available_bits_to_embed = 8
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed -= 1
_embedded += 1
else:
_thrown += 1
if self.n > 1:
try:
is_last_byte = False
filtered_index = FilteredCollection(permutation.shuffled[i+1:], lambda index: index % 64 and coeff[index])
while not is_last_byte:
k_bits_to_embed = 0
for i in range(k):
if available_bits_to_embed == 0:
if not self.embedded_data.available():
is_last_byte = True
break
byte_to_embed = self.embedded_data.read()
byte_to_embed ^= random.get_next_byte()
available_bits_to_embed = 8
next_bit_to_embed = byte_to_embed & 1
byte_to_embed >>= 1
available_bits_to_embed -= 1
k_bits_to_embed |= next_bit_to_embed << i
_embedded += 1
code_word = filtered_index.offer(self.n)
while True:
vhash = 0
for i, index in enumerate(code_word):
if coeff[index] > 0:
extracted_bit = coeff[index] & 1
else:
extracted_bit = 1 - (coeff[index] & 1)
if extracted_bit == 1:
vhash ^= i + 1
i = vhash ^ k_bits_to_embed
if not i:
break
i -= 1
coeff[code_word[i]] += 1 if coeff[code_word[i]] < 0 else -1
_changed += 1
if not coeff[code_word[i]]:
_thrown += 1
code_word[i:i+1] = []
code_word.extend(filtered_index.offer(1))
else:
break
except FilteredCollection.ListNotEnough:
pass
if _examined > 0:
logger.info('%d coefficients examined' % _examined)
if _changed > 0:
logger.info('%d coefficients changed (efficiency: %d.%d bits per change' % (_changed, _embedded / _changed, _embedded * 10 / _changed % 10))
logger.info('%d coefficients thrown (zeroed)' % _thrown)
logger.info('%d bits (%d bytes) embedded' % (_embedded, _embedded / 8))
logger.info('starting hufman encoding')
shuffled_index = 0
for r in range(min_block_height):
for c in range(min_block_width):
for comp in range(self.jpeg_obj.comp_num):
for i in range(self.jpeg_obj.vsamp_factor[comp]):
for j in range(self.jpeg_obj.hsamp_factor[comp]):
dct_array3 = coeff[shuffled_index:shuffled_index+64]
self.huf.huffman_block_encoder(self.out, dct_array3, last_dc_value[comp],
self.jpeg_obj.dctable_number[comp], self.jpeg_obj.actable_number[comp])
last_dc_value[comp] = dct_array3[0]
shuffled_index += 64
self.huf.flush_buffer(self.out)
