def handle_pushButton_cmos_snr_linearity_clicked(self, checked):
print("cmos_snr_linearity pushButton clicked", self)
i = 0
filePath = QFileDialog.getExistingDirectory(self.ui, "选择存储路径")
for root, dirs, files in os.walk(filePath):
for f in files:
i = i + 1 # 统计文件夹内总共有几个文件
print("i=", i)
if i < 2: # 少于2个文件，无法测试线性化程度
QMessageBox.warning(
self.ui,
'文件太少',
'文件夹内文件数目必须大于1个')
return
filename = [0] * i # 初始化成员为i个的一个列表
Rvalue = [0] * i
GRvalue = [0] * i
GBvalue = [0] * i
Bvalue = [0] * i
width = int(self.ui.lineEdit_cmos_snr_width.text())
height = int(self.ui.lineEdit_cmos_snr_height.text())
w_percent = int(self.ui.lineEdit_cmos_snr_w_percent.text()) / 100
h_percent = int(self.ui.lineEdit_cmos_snr_h_percent.text()) / 100
pattern = self.ui.comboBox_cmos_snr_cfa.currentText()
dataformat = self.ui.comboBox_cmos_snr_dataformat.currentText()
shift_bits = int(self.ui.comboBox_cmos_snr_bitshift.currentText())
WidthBorder = round((1 - w_percent) * width / 4)
HeightBorder = round((1 - h_percent) * height / 4)
# print("WidthBorder:", WidthBorder, HeightBorder, (width / 2 - WidthBorder), (height / 2 - HeightBorder))
i = 0
for root, dirs, files in os.walk(filePath):
for f in files:
filename[i] = os.path.join(root, f) # 将文件名填写到列表中
iamge = rawimage.read_plained_file(filename[i], "uint16", width, height, dataformat)
R, GR, GB, B, G = rawimage.bayer_channel_separation(iamge, pattern)
R = R[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
GR = GR[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
GB = GB[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
B = B[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
# print("shape:", np.shape(R))
Rvalue[i] = np.mean(R)
GRvalue[i] = np.mean(GR)
GBvalue[i] = np.mean(GB)
Bvalue[i] = np.mean(B)
print(filename[i])
i = i + 1
print("len = ", len(filename))
x = np.arange(0, i)
plt.plot(x, Rvalue, "r", label="R")
plt.plot(x, GRvalue, "g", label="GR")
plt.plot(x, GBvalue, "c", label="GB")
plt.plot(x, Bvalue, "b", label="B")
plt.title("Linearity")
plt.legend(loc="lower right")
plt.show()
self.ui.close()
def handle_pushButton_prnu_calculate_clicked(self, checked):
print("cmos_prnu_calculate pushButton clicked", self)
print("path1:\n", self.filePath1)
print("path2:\n", self.filePath2)
width = int(self.ui.lineEdit_prnu_width.text())
height = int(self.ui.lineEdit_prnu_height.text())
widthpercent = int(self.ui.lineEdit_prnu_widthpercent.text()) / 100
heightpercent = int(self.ui.lineEdit_prnu_heightpercent.text()) / 100
filtersize = int(self.ui.lineEdit_prnu_filter_size.text())
sensorbit = int(self.ui.comboBox_prnu_sensorbit.currentText())
pattern = self.ui.comboBox_prnu_cfa.currentText()
image1 = rawimage.read_plained_file(self.filePath1, 'uint16', width,
height, 'ieee-le')
image2 = rawimage.read_plained_file(self.filePath2, 'uint16', width,
height, 'ieee-le')
image = image1 - image2
R, GR, GB, B, G = rawimage.bayer_channel_separation(image, pattern)
WidthBorder = round((1 - widthpercent) * width / 4)
HeightBorder = round((1 - heightpercent) * height / 4)
G = G[HeightBorder:int(height / 2 - HeightBorder),
WidthBorder:int(width / 2 - WidthBorder)]
filter1 = np.ones(
(2 * filtersize + 1, 2 * filtersize + 1)) / ((2 * filtersize + 1) *
(2 * filtersize + 1))
filter_im = signal.convolve(G, filter1)
std_G = np.std(filter_im)
mean_G = np.mean(G)
prnu_value = 100 * std_G / mean_G
self.ui.lineEdit_prnu_value.setText(str(format(prnu_value, '.2f')))
def handle_pushButton_sensitivity_calculate_clicked(self, checked):
print("cmos_sensitivity_calculate pushButton clicked", self)
print("path1:\n", self.filePath1)
print("path2:\n", self.filePath2)
width = int(self.ui.lineEdit_sensitivity_width.text())
height = int(self.ui.lineEdit_sensitivity_height.text())
widthpercent = int(
self.ui.lineEdit_sensitivity_widthpercent.text()) / 100
heightpercent = int(
self.ui.lineEdit_sensitivity_heightpercent.text()) / 100
bright = int(self.ui.lineEdit_sensitivity_bright.text())
time_int = int(self.ui.lineEdit_sensitivity_time_int.text())
sensorbit = int(self.ui.comboBox_sensitivity_sensorbit.currentText())
pattern = self.ui.comboBox_sensitivity_cfa.currentText()
image_white = rawimage.read_plained_file(self.filePath1, 'uint16',
width, height, 'ieee-le')
image_black = rawimage.read_plained_file(self.filePath2, 'uint16',
width, height, 'ieee-le')
image = image_white - image_black
R, GR, GB, B, G = rawimage.bayer_channel_separation(image, pattern)
WidthBorder = round((1 - widthpercent) * width / 4)
HeightBorder = round((1 - heightpercent) * height / 4)
G = G[HeightBorder:int(height / 2 - HeightBorder),
WidthBorder:int(width / 2 - WidthBorder)]
G_mean = np.mean(G)
fsd = pow(2, sensorbit) - 1
sensitivity_value = G_mean / (bright * fsd / time_int)
self.ui.lineEdit_sensitivity_value.setText(
str(format(sensitivity_value, '.7f')))
def handle_pushButton_dsnu_calculate_clicked(self, checked):
print("cmos_dsnu_calculate pushButton clicked", self)
print("path1:\n", self.filePath1)
print("path2:\n", self.filePath2)
width = int(self.ui.lineEdit_dsnu_width.text())
height = int(self.ui.lineEdit_dsnu_height.text())
widthpercent = int(self.ui.lineEdit_dsnu_widthpercent.text()) / 100
heightpercent = int(self.ui.lineEdit_dsnu_heightpercent.text()) / 100
time_image1 = int(self.ui.lineEdit_dsnu_time_image1.text())
time_image2 = int(self.ui.lineEdit_dsnu_time_image2.text())
sensorbit = int(self.ui.comboBox_dsnu_sensorbit.currentText())
pattern = self.ui.comboBox_dsnu_cfa.currentText()
image1 = rawimage.read_plained_file(self.filePath1, 'uint16', width,
height, 'ieee-le')
image2 = rawimage.read_plained_file(self.filePath2, 'uint16', width,
height, 'ieee-le')
image = image1 - image2
R, GR, GB, B, G = rawimage.bayer_channel_separation(image, pattern)
WidthBorder = round((1 - widthpercent) * width / 4)
HeightBorder = round((1 - heightpercent) * height / 4)
G = G[HeightBorder:int(height / 2 - HeightBorder),
WidthBorder:int(width / 2 - WidthBorder)]
std_G = np.std(G)
fsd = pow(2, sensorbit) - 1
dsnu_value = 1000 * std_G / ((time_image1 - time_image2) * fsd)
self.ui.lineEdit_dsnu_value.setText(str(format(dsnu_value, '.7f')))
def bayer_average(image, pattern):
R, GR, GB, B = rawimage.bayer_channel_separation(image, pattern)
R_a = mono_average(R)
GR_a = mono_average(GR)
GB_a = mono_average(GB)
B_a = mono_average(B)
return R_a, GR_a, GB_a, B_a
def bayer_cumuhistogram(image, pattern, max1):
R, GR, GB, B = rawimage.bayer_channel_separation(image, pattern)
R_hist = mono_cumuhistogram(R, max1)
GR_hist = mono_cumuhistogram(GR, max1)
GB_hist = mono_cumuhistogram(GB, max1)
B_hist = mono_cumuhistogram(B, max1)
return R_hist, GR_hist, GB_hist, B_hist
def apply_shading_to_image(img, block_size, shading_R, shading_GR, shading_GB,
shading_B, pattern, ratio):
# 用G做luma
luma_shading = (shading_GR + shading_GB) / 2
# 计算color shading
R_color_shading = shading_R / luma_shading
GR_color_shading = shading_GR / luma_shading
GB_color_shading = shading_GB / luma_shading
B_color_shading = shading_B / luma_shading
# 计算调整之后luma shading
new_luma_shading = (luma_shading - 1) * ratio + 1
# 合并两种shading
new_shading_R = R_color_shading * new_luma_shading
new_shading_GR = GR_color_shading * new_luma_shading
new_shading_GB = GB_color_shading * new_luma_shading
new_shading_B = B_color_shading * new_luma_shading
R, GR, GB, B = rawimage.bayer_channel_separation(img, pattern)
HH, HW = R.shape
size_new = (HW + block_size, HH + block_size)
# 插值的方法的选择
ex_R_gain_map = cv.resize(new_shading_R,
size_new,
interpolation=cv.INTER_CUBIC)
ex_GR_gain_map = cv.resize(new_shading_GR,
size_new,
interpolation=cv.INTER_CUBIC)
ex_GB_gain_map = cv.resize(new_shading_GB,
size_new,
interpolation=cv.INTER_CUBIC)
ex_B_gain_map = cv.resize(new_shading_B,
size_new,
interpolation=cv.INTER_CUBIC)
# 裁剪到原图大小
half_b_size = int(block_size / 2)
R_gain_map = ex_R_gain_map[half_b_size:half_b_size + HH,
half_b_size:half_b_size + HW]
GR_gain_map = ex_GR_gain_map[half_b_size:half_b_size + HH,
half_b_size:half_b_size + HW]
GB_gain_map = ex_GB_gain_map[half_b_size:half_b_size + HH,
half_b_size:half_b_size + HW]
B_gain_map = ex_B_gain_map[half_b_size:half_b_size + HH,
half_b_size:half_b_size + HW]
R_new = R * R_gain_map
GR_new = GR * GR_gain_map
GB_new = GB * GB_gain_map
B_new = B * B_gain_map
new_image = rawimage.bayer_channel_integration(R_new, GR_new, GB_new,
B_new, pattern)
# 值缩减到0~1023
new_image = np.clip(new_image, a_min=0, a_max=1023)
return new_image
def raw_white_balance(image, type, sensorbit, pattern):
if sensorbit == 10:
smax = 1023
elif sensorbit == 12:
smax = 4095
else:
smax = 255
R, GR, GB, B, G = rawimage.bayer_channel_separation(image, pattern)
if type == "grey_world":
R_gain, G_gain, B_gain = grey_world(R, G, B)
elif type == "auto_threshold":
R_gain, G_gain, B_gain = auto_threshold(R, G, B)
elif type == "grey_world2":
R_gain, G_gain, B_gain = grey_edge(R, G, B, njet=0, mink_norm=1, sigma=0, saturation_threshold=smax)
elif type == "shade_of_grey":
R_gain, G_gain, B_gain = grey_edge(R, G, B, njet=0, mink_norm=5, sigma=0, saturation_threshold=smax)
elif type == "max_RGB":
R_gain, G_gain, B_gain = grey_edge(R, G, B, njet=0, mink_norm=-1, sigma=0, saturation_threshold=smax)
elif type == "grey_edge":
R_gain, G_gain, B_gain = grey_edge(R, G, B, njet=1, mink_norm=5, sigma=2, saturation_threshold=smax)
result_image = apply_raw(pattern, R, GR, GB, B, R_gain, G_gain, B_gain, smax)
# rawimage.show_planedraw(result_image, width=4032, height=2742, pattern="gray", sensorbit=10, compress_ratio=1)
# rawimage.show_planedraw(result_image, width=4032, height=2752, pattern="GRBG", sensorbit=10, compress_ratio=1)
h, w = R.shape
img = np.zeros(shape=(h, w, 3))
img2 = np.zeros(shape=(h, w, 3))
img[:, :, 0] = R
img[:, :, 1] = G
img[:, :, 2] = B
R2, GR2, GB2, B2, G2 = rawimage.bayer_channel_separation(result_image, pattern)
img2[:, :, 0] = R2
img2[:, :, 1] = GR2
img2[:, :, 2] = B2
rawimage.show_planedraw(img, w, h, pattern="color", sensorbit=10, compress_ratio=1)
rawimage.show_planedraw(img2, w, h, pattern="color", sensorbit=10, compress_ratio=1)
def apply_shading_to_image_base(img, block_size, shading_R, shading_GR,
shading_GB, shading_B, pattern, ratio):
# 用G做luma
luma_shading = (shading_GR + shading_GB) / 2
# 计算color shading
R_color_shading = shading_R / luma_shading
GR_color_shading = shading_GR / luma_shading
GB_color_shading = shading_GB / luma_shading
B_color_shading = shading_B / luma_shading
# 计算调整之后luma shading
new_luma_shading = (luma_shading - 1) * ratio + 1
# 合并两种shading
new_shading_R = R_color_shading * new_luma_shading
print("new_shading_R.shape:", new_shading_R.shape,
"R_color_shading.shape:", R_color_shading.shape,
"new_luma_shading.shape:", new_luma_shading.shape)
new_shading_GR = GR_color_shading * new_luma_shading
new_shading_GB = GB_color_shading * new_luma_shading
new_shading_B = B_color_shading * new_luma_shading
R, GR, GB, B = rawimage.bayer_channel_separation(img, pattern)
HH, HW = R.shape
size_new = (HW, HH) # 注意opencv和python的不同
# 插值的方法的选择
ex_R_gain_map = cv.resize(new_shading_R,
size_new,
interpolation=cv.INTER_CUBIC)
ex_GR_gain_map = cv.resize(new_shading_GR,
size_new,
interpolation=cv.INTER_CUBIC)
ex_GB_gain_map = cv.resize(new_shading_GB,
size_new,
interpolation=cv.INTER_CUBIC)
ex_B_gain_map = cv.resize(new_shading_B,
size_new,
interpolation=cv.INTER_CUBIC)
R_new = R * ex_R_gain_map
print("R_new.shape:", R_new.shape, "R.shape:", R.shape,
"ex_R_gain_map.shape:", ex_R_gain_map.shape)
GR_new = GR * ex_GR_gain_map
GB_new = GB * ex_GB_gain_map
B_new = B * ex_B_gain_map
new_image = rawimage.bayer_channel_integration(R_new, GR_new, GB_new,
B_new, pattern)
# 值缩减到0~1023
new_image = np.clip(new_image, a_min=0, a_max=1023)
return new_image
def handle_pushButton_temporal_noise_calculate_clicked(self, checked):
print("cmos_temporal_noise_calculate pushButton clicked", self)
i = 0
filePath = QFileDialog.getExistingDirectory(self.ui, "选择存储路径")
for root, dirs, files in os.walk(filePath):
for f in files:
i = i + 1 # 统计文件夹内总共有几个文件
# print("i=", i)
if i == 0: # 没有加载文件夹
return
img_num = i
filename = [0] * img_num # 初始化成员为i个的一个列表
width = int(self.ui.lineEdit_temporal_noise_width.text())
height = int(self.ui.lineEdit_temporal_noise_height.text())
roi_x = int(self.ui.lineEdit_temporal_noise_start_x.text())
roi_y = int(self.ui.lineEdit_temporal_noise_start_y.text())
roi_width = int(self.ui.lineEdit_temporal_noise_roi_width.text())
roi_height = int(self.ui.lineEdit_temporal_noise_roi_height.text())
bit_shift = int(self.ui.comboBox_temporal_noise_bitshift.currentText())
pattern = self.ui.comboBox_temporal_noise_cfa.currentText()
dataformat = self.ui.comboBox_temporal_noise_dataformat.currentText()
inputformat = self.ui.comboBox_temporal_noise_inputformat.currentText()
Rvalue = np.empty((img_num, roi_height, roi_width))
i = 0
for root, dirs, files in os.walk(filePath):
for f in files:
filename[i] = os.path.join(root, f) # 将文件名填写到列表中
image = rawimage.read_plained_file(filename[i], inputformat,
width, height, dataformat)
R, GR, GB, B, G = rawimage.bayer_channel_separation(
image, pattern)
Rvalue[i] = R[roi_y:(roi_y + roi_height),
roi_x:(roi_x + roi_width)]
i = i + 1
var_Rvalue = Rvalue.var(axis=0) # 不同图像同一位置的像素求方差
mean_std = math.sqrt(
np.sum(np.sum(var_Rvalue, axis=1), axis=0) /
(roi_width * roi_height))
fsd = pow(2, (16 + bit_shift)) - 1
db = 20 * np.log10(mean_std / fsd)
db = format(db, '.2f')
mean_std = format(mean_std, '.2f')
self.ui.lineEdit_temporal_noise_emva1288.setText(str(mean_std))
self.ui.lineEdit_temporal_noise_smia.setText(str(db))
def binning_image(image, height, width, block_size_h, block_size_w, pattern):
region_h_n = int(height / block_size_h)
region_w_n = int(width / block_size_w)
binning_image1 = np.empty((region_h_n * 2, region_w_n * 2),
dtype=np.float32)
x1 = 0
y1 = 0
for j in range(region_h_n):
for i in range(region_w_n):
region_data = get_region(image, y1, x1, block_size_h, block_size_w)
R, GR, GB, B = rawimage.bayer_channel_separation(
region_data, pattern)
binning_image1[j * 2, i * 2] = np.mean(R)
binning_image1[j * 2, (i * 2) + 1] = np.mean(GR)
binning_image1[(j * 2) + 1, i * 2] = np.mean(GB)
binning_image1[(j * 2) + 1, (i * 2) + 1] = np.mean(B)
x1 = x1 + block_size_w
y1 = y1 + block_size_h
x1 = 0
return binning_image1
def handle_pushButton_cmos_snr_snr_clicked(self, checked):
print("cmos_snr_snr pushButton clicked", self)
i = 0
filePath = QFileDialog.getExistingDirectory(self.ui, "选择存储路径")
for root, dirs, files in os.walk(filePath):
for f in files:
i = i + 1 # 统计文件夹内总共有几个文件
print("i=", i)
if i == 0: # 没有加载文件夹
return
filename = [0] * i # 初始化成员为i个的一个列表
Rvalue = [0] * i
RNoise = [0] * i
R_SNR = [0] * i
GRvalue = [0] * i
GRNoise = [0] * i
GR_SNR = [0] * i
GBvalue = [0] * i
GBNoise = [0] * i
GB_SNR = [0] * i
Bvalue = [0] * i
BNoise = [0] * i
B_SNR = [0] * i
width = int(self.ui.lineEdit_cmos_snr_width.text())
height = int(self.ui.lineEdit_cmos_snr_height.text())
w_percent = int(self.ui.lineEdit_cmos_snr_w_percent.text()) / 100
h_percent = int(self.ui.lineEdit_cmos_snr_h_percent.text()) / 100
pattern = self.ui.comboBox_cmos_snr_cfa.currentText()
dataformat = self.ui.comboBox_cmos_snr_dataformat.currentText()
shift_bits = int(self.ui.comboBox_cmos_snr_bitshift.currentText())
WidthBorder = round((1 - w_percent) * width / 4)
HeightBorder = round((1 - h_percent) * height / 4)
# print("WidthBorder:", WidthBorder, HeightBorder, (width / 2 - WidthBorder), (height / 2 - HeightBorder))
i = 0
for root, dirs, files in os.walk(filePath):
for f in files:
filename[i] = os.path.join(root, f) # 将文件名填写到列表中
iamge = rawimage.read_plained_file(filename[i], "uint16", width, height, dataformat)
R, GR, GB, B, G = rawimage.bayer_channel_separation(iamge, pattern)
R = R[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
GR = GR[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
GB = GB[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
B = B[HeightBorder:int(height / 2 - HeightBorder), WidthBorder:int(width / 2 - WidthBorder)]
# print("shape:", np.shape(R))
Rvalue[i] = np.mean(R)
RNoise[i] = np.std(R)
R_SNR[i] = Rvalue[i] / RNoise[i]
GRvalue[i] = np.mean(GR)
GRNoise[i] = np.std(GR)
GR_SNR[i] = GRvalue[i] / GRNoise[i]
GBvalue[i] = np.mean(GB)
GBNoise[i] = np.std(GB)
GB_SNR[i] = GBvalue[i] / GBNoise[i]
Bvalue[i] = np.mean(B)
BNoise[i] = np.std(B)
B_SNR[i] = Bvalue[i] / BNoise[i]
print(filename[i])
i = i + 1
print("len = ", len(filename))
if i > 1:
x = np.arange(0, i)
plt.plot(x, R_SNR, "r", label="R")
plt.plot(x, GR_SNR, "g", label="GR")
plt.plot(x, GB_SNR, "c", label="GB")
plt.plot(x, B_SNR, "b", label="B")
else:
plt.scatter(1, R_SNR[0], color="r", label="R", linewidth=3)
plt.scatter(2, GR_SNR[0], color="g", label="GR", linewidth=3)
plt.scatter(3, GB_SNR[0], color="c", label="GB", linewidth=3)
plt.scatter(4, B_SNR[0], color="b", label="B", linewidth=3)
plt.title("SNR")
plt.legend(loc="lower right")
plt.show()
self.ui.close()
def handle_pushButton_fpn_calculate_clicked(self, checked):
print("cmos_fpn_calculate pushButton clicked", self)
i = 0
filePath = QFileDialog.getExistingDirectory(self.ui, "选择存储路径")
for root, dirs, files in os.walk(filePath):
for f in files:
i = i + 1 # 统计文件夹内总共有几个文件
# print("i=", i)
if i == 0: # 没有加载文件夹
return
img_num = i
filename = [0] * img_num # 初始化成员为i个的一个列表
width = int(self.ui.lineEdit_fpn_width.text())
height = int(self.ui.lineEdit_fpn_height.text())
roi_x = int(self.ui.lineEdit_fpn_start_x.text())
roi_y = int(self.ui.lineEdit_fpn_start_y.text())
roi_width = int(self.ui.lineEdit_fpn_roi_width.text())
roi_height = int(self.ui.lineEdit_fpn_roi_height.text())
bit_shift = int(self.ui.comboBox_fpn_bitshift.currentText())
pattern = self.ui.comboBox_fpn_cfa.currentText()
dataformat = self.ui.comboBox_fpn_dataformat.currentText()
inputformat = self.ui.comboBox_fpn_inputformat.currentText()
Rvalue = np.zeros((roi_height, roi_width), np.int)
average_Rvalue = np.zeros((roi_height, roi_width), np.int)
i = 0
for root, dirs, files in os.walk(filePath):
for f in files:
filename[i] = os.path.join(root, f) # 将文件名填写到列表中
image = rawimage.read_plained_file(filename[i], inputformat,
width, height, dataformat)
R, GR, GB, B, G = rawimage.bayer_channel_separation(
image, pattern)
Rvalue = R[roi_y:(roi_y + roi_height),
roi_x:(roi_x + roi_width)]
average_Rvalue = average_Rvalue + Rvalue
# print("Rvalue = \n", Rvalue[0][0])
# print("average_Rvalue = \n", average_Rvalue[0][0])
i = i + 1
average_Rvalue = average_Rvalue / i
# print("shape = ", average_Rvalue)
Noise_tol = np.std(Rvalue) # 所有元素参加计算标准差
FPN_total = np.std(average_Rvalue)
# print("Noise_tol = ", Noise_tol, str(Noise_tol))
# print("FPN_total = ", FPN_total, str(FPN_total))
self.ui.lineEdit_fpn_total.setText(str(format(FPN_total, '.7f')))
self.ui.lineEdit_fpn_temporal_noise.setText(
str(format(Noise_tol - FPN_total, '.7f')))
column_array = np.mean(average_Rvalue, axis=0) # axis=0，计算每一列的均值
row_array = np.mean(average_Rvalue, axis=1) # axis=1，计算每一行的均值
sum_c = 0
sum_r = 0
# print("shape:", column_array.shape[0], row_array.shape)
for i in range(0, column_array.shape[0] - 1):
sum_c = sum_c + pow((column_array[i] - column_array[i + 1]), 2)
for i in range(0, row_array.shape[0] - 1):
sum_r = sum_r + pow((row_array[i] - row_array[i + 1]), 2)
fsd = pow(2, (16 + bit_shift)) - 1
fpn_v_level = np.sqrt(sum_c / (column_array.shape[0] - 1)) / fsd
fpn_h_level = np.sqrt(sum_r / (row_array.shape[0] - 1)) / fsd
KERNEL = np.array([-1, -1, -1, -1, -1, 10, -1, -1, -1, -1, -1]) / 10
fpn_v_max = max(signal.convolve(column_array, KERNEL)) / fsd
fpn_h_max = max(signal.convolve(row_array, KERNEL)) / fsd
self.ui.lineEdit_fpn_v_level.setText(str(format(fpn_v_level, '.7f')))
self.ui.lineEdit_fpn_h_level.setText(str(format(fpn_h_level, '.7f')))
self.ui.lineEdit_fpn_v_max.setText(str(format(fpn_v_max, '.7f')))
self.ui.lineEdit_fpn_h_max.setText(str(format(fpn_h_max, '.7f')))
def create_lsc_data(img, block_size, pattern):
# 分开四个颜色通道
R, GR, GB, B = rawimage.bayer_channel_separation(img, pattern)
# print(img.shape, R.shape)
# 每张的高宽
HH, HW = R.shape
# 生成分多少块
Hblocks = int(HH / block_size)
Wblocks = int(HW / block_size)
# 整个图像被分成 Hblocks * Wblocks 块，生成一个 Hblocks * Wblocks 的矩阵，用于数据储存。
R_LSC_data = np.zeros((Hblocks, Wblocks)) # 每个块 R 的平均值
B_LSC_data = np.zeros((Hblocks, Wblocks))
GR_LSC_data = np.zeros((Hblocks, Wblocks))
GB_LSC_data = np.zeros((Hblocks, Wblocks))
# 块距离光心的距离
RA = np.zeros((Hblocks, Wblocks))
# 计算每个块的平均值
for y in range(0, HH, block_size):
for x in range(0, HW, block_size):
block_y_num = int(y / block_size)
block_x_num = int(x / block_size)
R_LSC_data[block_y_num,
block_x_num] = R[y:y + block_size, x:x +
block_size].mean() # 对一块block数据求平均
GR_LSC_data[block_y_num,
block_x_num] = GR[y:y + block_size,
x:x + block_size].mean()
GB_LSC_data[block_y_num,
block_x_num] = GB[y:y + block_size,
x:x + block_size].mean()
B_LSC_data[block_y_num, block_x_num] = B[y:y + block_size,
x:x + block_size].mean()
# 根据GR的最大值，寻找真正的光心块
center_point = np.where(GR_LSC_data == np.max(GR_LSC_data))
center_y = center_point[0] * block_size + block_size / 2
center_x = center_point[1] * block_size + block_size / 2
# 计算块距离光心的距离
for y in range(0, HH, block_size):
for x in range(0, HW, block_size):
xx = x + block_size / 2
yy = y + block_size / 2
block_y_num = int(y / block_size)
block_x_num = int(x / block_size)
RA[block_y_num, block_x_num] = (yy - center_y) * (
yy - center_y) + (xx - center_x) * (xx - center_x)
# 4个颜色数据通道展平，便于数据进行拟合，RA_flatten相当于x，R_LSC_data_flatten和其他三个相当于y
RA_flatten = RA.flatten()
R_LSC_data_flatten = R_LSC_data.flatten()
GR_LSC_data_flatten = GR_LSC_data.flatten()
GB_LSC_data_flatten = GB_LSC_data.flatten()
B_LSC_data_flatten = B_LSC_data.flatten()
# 最亮块的值
Max_R = np.max(R_LSC_data_flatten)
Max_GR = np.max(GR_LSC_data_flatten)
Max_GB = np.max(GB_LSC_data_flatten)
Max_B = np.max(B_LSC_data_flatten)
# 得到gain,还没有外插
G_R_LSC_data = Max_R / R_LSC_data_flatten
G_GR_LSC_data = Max_GR / GR_LSC_data_flatten
G_GB_LSC_data = Max_GB / GB_LSC_data_flatten
G_B_LSC_data = Max_B / B_LSC_data_flatten
# gain
plt.scatter(RA_flatten, G_R_LSC_data, color='red')
plt.scatter(RA_flatten, G_GR_LSC_data, color='green')
plt.scatter(RA_flatten, G_GB_LSC_data, color='green')
plt.scatter(RA_flatten, G_B_LSC_data, color='blue')
plt.show()
# 进行gain曲线拟合拟合
par_R = np.polyfit(RA_flatten, G_R_LSC_data, 3)
par_GR = np.polyfit(RA_flatten, G_GR_LSC_data, 3)
par_GB = np.polyfit(RA_flatten, G_GB_LSC_data, 3)
par_B = np.polyfit(RA_flatten, G_B_LSC_data, 3)
# 拟合之后生成所有点的值
ES_R = par_R[0] * (RA_flatten**3) + par_R[1] * (
RA_flatten**2) + par_R[2] * RA_flatten + par_R[3]
ES_GR = par_GR[0] * (RA_flatten**3) + par_GR[1] * (
RA_flatten**2) + par_GR[2] * RA_flatten + par_GR[3]
ES_GB = par_GB[0] * (RA_flatten**3) + par_GB[1] * (
RA_flatten**2) + par_GB[2] * RA_flatten + par_GB[3]
ES_B = par_B[0] * (RA_flatten**3) + par_B[1] * (
RA_flatten**2) + par_B[2] * RA_flatten + par_B[3]
# 拟合数据和原有数据有什么不同
plt.scatter(RA_flatten, ES_R, color='red')
plt.scatter(RA_flatten, ES_GR, color='green')
plt.scatter(RA_flatten, ES_GB, color='green')
plt.scatter(RA_flatten, ES_B, color='blue')
plt.show()
# 外插补偿的gain通过曲线拟合得到，这边进行外插主要是考虑有些图像的分辨率不能被block整除
EX_RA = np.zeros((Hblocks + 2, Wblocks + 2))
EX_R = np.zeros((Hblocks + 2, Wblocks + 2))
EX_GR = np.zeros((Hblocks + 2, Wblocks + 2))
EX_GB = np.zeros((Hblocks + 2, Wblocks + 2))
EX_B = np.zeros((Hblocks + 2, Wblocks + 2))
new_center_y = center_point[0] + 1
new_center_x = center_point[1] + 1
for y in range(0, Hblocks + 2):
for x in range(0, Wblocks + 2):
EX_RA[y, x] = (y - new_center_y) * block_size * (
y - new_center_y) * block_size + (
x - new_center_x) * block_size * (
x - new_center_x) * block_size
EX_R[y, x] = par_R[0] * (EX_RA[y, x] ** 3) + par_R[1] * (EX_RA[y, x] ** 2) + par_R[2] * (EX_RA[y, x]) + \
par_R[3]
EX_GR[y, x] = par_GR[0] * (EX_RA[y, x] ** 3) + par_GR[1] * (EX_RA[y, x] ** 2) + par_GR[2] * (EX_RA[y, x]) + \
par_GR[3]
EX_GB[y, x] = par_GB[0] * (EX_RA[y, x] ** 3) + par_GB[1] * (EX_RA[y, x] ** 2) + par_GB[2] * (EX_RA[y, x]) + \
par_GB[3]
EX_B[y, x] = par_B[0] * (EX_RA[y, x] ** 3) + par_B[1] * (EX_RA[y, x] ** 2) + par_B[2] * (EX_RA[y, x]) + \
par_B[3]
# 中心用实际采样的数据
G_R_LSC_data.shape = (Hblocks, Wblocks)
G_GR_LSC_data.shape = (Hblocks, Wblocks)
G_GB_LSC_data.shape = (Hblocks, Wblocks)
G_B_LSC_data.shape = (Hblocks, Wblocks)
EX_R[1:1 + Hblocks, 1:1 + Wblocks] = G_R_LSC_data
EX_GR[1:1 + Hblocks, 1:1 + Wblocks] = G_GR_LSC_data
EX_GB[1:1 + Hblocks, 1:1 + Wblocks] = G_GB_LSC_data
EX_B[1:1 + Hblocks, 1:1 + Wblocks] = G_B_LSC_data
return EX_R, EX_GR, EX_GB, EX_B
def create_lsc_data_base(img, block_size, pattern):
# 分开四个颜色通道
R, GR, GB, B = rawimage.bayer_channel_separation(img, pattern)
# print(img.shape, R.shape)
# 每张的高宽
HH, HW = R.shape
# 生成分多少块
Hblocks = int(HH / block_size)
Wblocks = int(HW / block_size)
# 整个图像被分成 Hblocks * Wblocks 块，生成一个 Hblocks * Wblocks 的矩阵，用于数据储存。
R_LSC_data = np.zeros((Hblocks, Wblocks)) # 每个块 R 的平均值
B_LSC_data = np.zeros((Hblocks, Wblocks))
GR_LSC_data = np.zeros((Hblocks, Wblocks))
GB_LSC_data = np.zeros((Hblocks, Wblocks))
# 块距离光心的距离
RA = np.zeros((Hblocks, Wblocks))
# 计算每个块的平均值
for y in range(0, HH, block_size):
for x in range(0, HW, block_size):
block_y_num = int(y / block_size)
block_x_num = int(x / block_size)
R_LSC_data[block_y_num,
block_x_num] = R[y:y + block_size, x:x +
block_size].mean() # 对一块block数据求平均
GR_LSC_data[block_y_num,
block_x_num] = GR[y:y + block_size,
x:x + block_size].mean()
GB_LSC_data[block_y_num,
block_x_num] = GB[y:y + block_size,
x:x + block_size].mean()
B_LSC_data[block_y_num, block_x_num] = B[y:y + block_size,
x:x + block_size].mean()
# 4个颜色数据通道展平，便于数据进行拟合，RA_flatten相当于x，R_LSC_data_flatten和其他三个相当于y
R_LSC_data_flatten = R_LSC_data.flatten()
GR_LSC_data_flatten = GR_LSC_data.flatten()
GB_LSC_data_flatten = GB_LSC_data.flatten()
B_LSC_data_flatten = B_LSC_data.flatten()
# 最亮块的值
Max_R = np.max(R_LSC_data_flatten)
Max_GR = np.max(GR_LSC_data_flatten)
Max_GB = np.max(GB_LSC_data_flatten)
Max_B = np.max(B_LSC_data_flatten)
# 得到gain
G_R_LSC_data = Max_R / R_LSC_data_flatten
G_GR_LSC_data = Max_GR / GR_LSC_data_flatten
G_GB_LSC_data = Max_GB / GB_LSC_data_flatten
G_B_LSC_data = Max_B / B_LSC_data_flatten
G_R_LSC_data.shape = (Hblocks, Wblocks)
G_GR_LSC_data.shape = (Hblocks, Wblocks)
G_GB_LSC_data.shape = (Hblocks, Wblocks)
G_B_LSC_data.shape = (Hblocks, Wblocks)
return G_R_LSC_data, G_GR_LSC_data, G_GB_LSC_data, G_B_LSC_data
