def __init__(self, filename, image, parent=None):
super(QualityWidget, self).__init__(parent)
x = np.arange(1, 101)
y = loss_curve(image)
tail = 5
qm = np.argmin(y[:-tail]) + 1
if qm == 100 - tail:
qm = 100
figure = Figure()
canvas = FigureCanvas(figure)
axes = canvas.figure.subplots()
axes.plot(x, y * 100, label="compression loss")
axes.fill_between(x, y * 100, alpha=0.2)
axes.axvline(qm,
linestyle=":",
color="k",
label=f"min error (q = {qm})")
xt = axes.get_xticks()
xt = np.append(xt, 1)
axes.set_xticks(xt)
axes.set_xlim([1, 100])
axes.set_ylim([0, 100])
axes.set_xlabel(self.tr("JPEG quality (%)"))
axes.set_ylabel(self.tr("average error (%)"))
axes.grid(True, which="both")
axes.legend(loc="upper center")
axes.figure.canvas.draw()
figure.set_tight_layout(True)
main_layout = QVBoxLayout()
main_layout.addWidget(canvas)
MRK = b"\xFF"
SOI = b"\xD8"
DQT = b"\xDB"
# DHT = b'\xC4'
MSK = b"\x0F"
PAD = b"\x00"
MAX_TABLES = 2
LEN_OFFSET = 2
LUMA_IDX = 0
CHROMA_IDX = 1
luma = np.zeros((DCT_SIZE, DCT_SIZE), dtype=int)
chroma = np.zeros((DCT_SIZE, DCT_SIZE), dtype=int)
temp_file = QTemporaryFile()
try:
if temp_file.open():
copyfile(filename, temp_file.fileName())
subprocess.run(
[
exiftool_exe(), "-all=", "-overwrite_original",
temp_file.fileName()
],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
found = False
with open(temp_file.fileName(), "rb") as file:
first = file.read(1)
if first not in [MRK, SOI]:
raise ValueError(self.tr("File is not a JPEG image!"))
while True:
if not self.find_next(file, [MRK, DQT, PAD]):
break
length = file.read(1)[0] - LEN_OFFSET
if length <= 0 or length % (TABLE_SIZE + 1) != 0:
continue
while length > 0:
mode = file.read(1)
if not mode:
break
index = mode[0] & MSK[0]
if index >= MAX_TABLES:
break
length -= 1
for k in range(TABLE_SIZE):
b = file.read(1)[0]
if not b:
break
length -= 1
i, j = ZIG_ZAG[k]
if index == LUMA_IDX:
luma[i, j] = b
elif index == CHROMA_IDX:
chroma[i, j] = b
else:
found = True
if not found:
raise ValueError(self.tr("Unable to find JPEG tables!"))
levels = [(1 - (np.mean(t.ravel()[1:]) - 1) / 254) * 100
for t in [luma, chroma]]
distance = np.zeros(101)
for qm in range(101):
lu, ch = cv.split(get_tables(qm))
lu_diff = np.mean(cv.absdiff(luma, lu))
ch_diff = np.mean(cv.absdiff(chroma, ch))
distance[qm] = (lu_diff + 2 * ch_diff) / 3
closest = np.argmin(distance)
deviation = distance[closest]
if deviation == 0:
quality = closest
message = "(standard tables)"
else:
quality = int(np.round(closest - deviation))
message = f"(deviation from standard tables = {deviation:.4f})"
if quality == 0:
quality = 1
quality_label = QLabel(
self.tr(
f"[JPEG FORMAT] Last saved quality: {quality}% {message}"))
modify_font(quality_label, bold=True)
luma_label = QLabel(
self.
tr(f"Luminance Quantization Table (level = {levels[0]:.2f}%)\n"
))
luma_label.setAlignment(Qt.AlignCenter)
modify_font(luma_label, underline=True)
luma_table = self.create_table(luma)
luma_table.setFixedSize(420, 190)
chroma_label = QLabel(
self.
tr(f"Chrominance Quantization Table (level = {levels[1]:.2f}%)\n"
))
chroma_label.setAlignment(Qt.AlignCenter)
modify_font(chroma_label, underline=True)
chroma_table = self.create_table(chroma)
chroma_table.setFixedSize(420, 190)
table_layout = QGridLayout()
table_layout.addWidget(luma_label, 0, 0)
table_layout.addWidget(luma_table, 1, 0)
table_layout.addWidget(chroma_label, 0, 1)
table_layout.addWidget(chroma_table, 1, 1)
table_layout.addWidget(quality_label, 2, 0, 1, 2)
main_layout.addLayout(table_layout)
except ValueError:
modelfile = "models/jpeg_qf.mdl"
try:
model = load(modelfile)
limit = model.best_ntree_limit if hasattr(
model, "best_ntree_limit") else None
# f = self.get_features(image)
# p = model.predict_proba(f, ntree_limit=limit)[0, 0]
qp = model.predict(np.reshape(y, (1, len(y))),
ntree_limit=limit)[0]
# if p > 0.5:
# p = 2 * (p - 0.5) * 100
# output = self.tr('Uncompressed image (p = {:.2f}%)'.format(p))
# else:
# p = (1 - 2 * p) * 100
# output = self.tr('Compressed image (p = {:.2f}%) ---> Estimated JPEG quality = {}%'.format(p, qm))
message = self.tr(
f"[LOSSLESS FORMAT] Estimated last saved quality = {qp:.1f}%{'' if qp <= 99 else ' (uncompressed)'}"
)
if qp == 100:
message += " (uncompressed)"
prob_label = QLabel(message)
modify_font(prob_label, bold=True)
main_layout.addWidget(prob_label)
except FileNotFoundError:
QMessageBox.critical(
self, self.tr("Error"),
self.tr(f'Model not found ("{modelfile}")!'))
main_layout.addStretch()
self.setLayout(main_layout)
def get_features(image):
# q = [6, 8, 10, 13, 31, 71, 73, 75, 76, 81, 84, 87, 88, 93, 94, 96, 97, 98, 99, 100]
q = list(range(1, 101))
c = loss_curve(image, q)
return np.reshape(c, (1, len(q)))