def __init__(self, iplimage):
# Rough-n-ready but it works dammit
# alpha = cv.CreateMat(iplimage.height,iplimage.width, cv.CV_8UC1)
alpha = np.zeros((iplimage.shape[0], iplimage.shape[1]), np.uint8)
# Zieht ein schwarzes Rechteck ueber das Bild
cv2.rectangle(alpha, (0, 0), (iplimage.shape[1], iplimage.shape[0]),
cv.ScalarAll(255), -1)
rgba = np.zeros((iplimage.shape[0], iplimage.shape[1], 4), np.uint8)
#cv2.Set(rgba, (1, 2, 3, 4))
cv2.mixChannels(
[iplimage, alpha],
[rgba],
[
0,
0, # rgba[0] -> bgr[2]
1,
1, # rgba[1] -> bgr[1]
2,
2, # rgba[2] -> bgr[0]
3,
3 # rgba[3] -> alpha[0]
])
self.__imagedata = rgba.tostring()
super(IplQImage,
self).__init__(self.__imagedata, iplimage.shape[1],
iplimage.shape[0], QtGui.QImage.Format_RGB32)
def Hist_and_Backproj(val, target, hue):
## [initialize]
bins = val
histSize = max(bins, 2)
ranges = [0, 180] # hue_range
## [initialize]
hsvt = cv.cvtColor(target, cv.COLOR_BGR2HSV)
ch = (0, 0)
hue2 = np.empty(hsvt.shape, hsvt.dtype)
cv.mixChannels([hsvt], [hue2], ch)
hist2 = cv.calcHist([hue], [0], None, [histSize], ranges, accumulate=False)
hist = cv.calcHist([hue], [0], None, [histSize], ranges, accumulate=False)
cv.normalize(hist, hist, alpha=0, beta=255, norm_type=cv.NORM_MINMAX)
backproj = cv.calcBackProject([hue2], [0], hist, ranges, scale=1)
w = 400
h = 400
bin_w = int(round(w / histSize))
histImg = np.zeros((h, w, 3), dtype=np.uint8)
for i in range(bins):
cv.rectangle(histImg, (i * bin_w, h),
((i + 1) * bin_w, h - int(round(hist[i] * h / 255.0))),
(0, 0, 255), cv.FILLED)
cv.imshow('Histogram', histImg)
return backproj
def main(argv):
global src
src = cv2.imread(sys.argv[1], cv2.IMREAD_COLOR)
if src is None:
print 'Usage:\ncalc_back_project.py <path_to_image>'
return -1
global hsv
hsv = cv2.cvtColor(src, cv2.COLOR_BGR2HSV)
global hue
hue = np.zeros(hsv.shape, dtype=np.uint8)
ch = [0] * 2
cv2.mixChannels([hsv], [hue], ch)
window_image = 'Source image'
cv2.namedWindow(window_image, cv2.WINDOW_AUTOSIZE)
cv2.createTrackbar('* Hue bins: ', window_image, bins, 180,
Hist_and_Backproj)
Hist_and_Backproj(bins)
cv2.imshow(window_image, src)
cv2.waitKey(0)
return 0
def __init__(self, drawingOverlay, last_gframe):
# convert red pixels to yellow for the overlay
self.drawingOverlay = np.zeros_like(drawingOverlay)
cv2.mixChannels(
[drawingOverlay],
[self.drawingOverlay],
(0, 0, 2, 1, 2, 2)
)
self.shape = (drawingOverlay.shape[1], drawingOverlay.shape[0])
# Grab all pixels of overlay
xs, ys, zs = np.where(drawingOverlay > 0)
self.overlayPts = zip(xs, ys)
# Grab bounding rect
ptsMat = np.float32(map(lambda x: [x], self.overlayPts))
y, x, h, w = cv2.boundingRect(ptsMat)
x1, y1 = x, y
x2, y2 = x + w, y + h
# Create dense optical flow tracker
self.first_flow_pts = np.float32([
[x1, y1],
[x2, y1],
[x2, y2],
[x1, y2]
])
self.flow_tracker = OpticalFlowHomographyTracker(
last_gframe,
self.first_flow_pts
)
def find_squares(image):
gray0=cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
pyr=cv2.pyrDown(gray0)
timg=cv2.pyrUp(pyr)
for c in range (0,3):
cv2.mixChannels([timg], [gray0], [0,0])
for l in range (0,N):
if l == 0:
gray=cv2.Canny(gray0,thresh,100)
kernel = np.ones((5,5),np.uint8)
dilation = cv2.dilate(gray,kernel,iterations = 2)
else:
gray = gray0 >= (l+1)*255/N;
ret, contours, hire=cv2.findContours(dilation,cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
for i in range (0,len(contours)):
approx= cv2.approxPolyDP(contours[i],3,1)
convex = cv2.isContourConvex(contours[i])
if len(approx)==4 and math.fabs(cv2.contourArea(contours[i])) and convex== True:
print("square")
maxCosine = 0
for j in range(2,5):
cosine = math.fabs(angle(approx[j%4], approx[j-2], approx[j-1]));
if (maxCosine>cosine):
maxCosine=cosine
if maxCosine < 0.3:
squares.append(approx)
def switch_by_mixchannels(src):
"""
:param src:
:return:
"""
dst = np.zeros_like(src)
cv2.mixChannels([src], [dst], fromTo=[0, 2, 1, 1, 2, 0])
return dst
def processFrames(socket, frameWidth, frameHeight):
global dlibObject, counter, svc
frameSize = frameWidth * frameHeight * 4 # assuming ARGB
while True:
timestamp, frameNumber, frameName, frame, err = readFrame(
socket, frameSize)
if not err:
print(" > read frame " + str(frameNumber) + ": " + frameName +
" (" + str(len(frame)) + " bytes)")
imgARGB = np.frombuffer(frame,
'uint8').reshape(frameHeight, frameWidth,
4)
imgAlpha = np.zeros((frameHeight, frameWidth, 1), 'uint8')
imgBGR = np.zeros((frameHeight, frameWidth, 3), 'uint8')
cv2.mixChannels([imgARGB], [imgBGR, imgAlpha],
[1, 2, 2, 1, 3, 0, 0, 3])
# cv2.imwrite('test-frame'+str(counter)+'.jpg', imgBGR)
# counter += 1
# continue
# sys.exit(0)
p1 = datetime.datetime.now()
rects = dlibObject.getAllFaceBoundingBoxes(imgBGR)
if len(rects) > 0:
print(" > DETECTED " + str(len(rects)) + " faces")
facesArray = []
for r in rects:
faceAnnotation = getAnnotationFromRect(
r, frameWidth, frameHeight)
# run classifier, if probabilirt less than 50%, ignore classifier's data
faceAnnotation = runClassifier(imgBGR, faceAnnotation, r)
p2 = datetime.datetime.now()
delta = p2 - p1
processingMs = int(delta.total_seconds() * 1000)
print(" > open face processing took " + str(processingMs) +
" ms")
facesArray.append(faceAnnotation)
drawBox(imgBGR, r, str(faceAnnotation['label']))
dumpAnnotations(timestamp, frameNumber, frameName, facesArray)
else:
print(" > no faces detected")
dumpImage(imgBGR)
# handy code to slice image into separate channels
# imgR = np.zeros((frameHeight, frameWidth, 1), 'uint8')
# imgG = np.zeros((frameHeight, frameWidth, 1), 'uint8')
# imgB = np.zeros((frameHeight, frameWidth, 1), 'uint8')
# imgA = np.zeros((frameHeight, frameWidth, 1), 'uint8')
# cv2.mixChannels([imgARGB], [imgA, imgR, imgG, imgB], [0,0, 1,1, 2,2, 3,3])
# cv2.imwrite('test-frame-A.jpg', imgA)
# cv2.imwrite('test-frame-R.jpg', imgR)
# cv2.imwrite('test-frame-G.jpg', imgG)
# cv2.imwrite('test-frame-B.jpg', imgB)
else:
print(" > error reading frame: " + str(err))
def swap_channels(self, channels):
image_format = self.image_format
if ((channels is ImageFormat.BGR and image_format.channels is ImageFormat.RGB) or
(channels is ImageFormat.RGB and image_format.channels is ImageFormat.BGR)):
output = self.__class__(image_format, channels=channels)
cv2.mixChannels([self], [output], (0,2, 1,1, 2,0))
return output
# ???
else:
raise NotImplementedError
def on_btn2_3_click(self):
img = cv2.imread("../images/q2_train.jpg")
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
hue = np.zeros(hsv.shape, dtype=np.uint8)
cv2.mixChannels([hsv], [hue], [0, 0])
hist = cv2.calcHist([img], [0], None, [256], [0, 256])
# hist = cv2.calcHist([hue], [0], None, [256], [0,256])
cv2.normalize(hist, hist, 0, 255, cv2.NORM_MINMAX, -1)
backproj = cv2.calcBackProject([hue], [0, 1], hist, [0, 256], 1)
cv2.imshow("backproj", backproj)
cv2.waitKey(0)
cv2.destroyAllWindows()
def backproject():
srcImg = cv.imread(img_path + 'hu.jpg')
hsvImg = cv.cvtColor(srcImg, cv.COLOR_BGR2HSV)
hueImg = np.zeros(hsvImg.shape, np.uint8)
hueImg = cv.mixChannels([hsvImg], [hueImg], [0, 0])[0] # 取数值的第一个
# 调整值
def on_BinChg(v):
# 计算直方图
if v < 2:
v = 2
# sample
hist = cv.calcHist([hueImg], [0], None, [v], [0, 180])
hist = cv.normalize(hist, hist, 0, 255, cv.NORM_MINMAX, -1)
# target
backImg = cv.calcBackProject([hueImg], [0], hist, [0, 180], 1)
cv.imshow('BackImg', backImg)
# 绘制直方图
histImg = np.zeros((400, 400, 3), np.uint8)
bin_w = int(400 / v)
for i in range(v):
cv.rectangle(histImg, (i * bin_w, 400),
((i + 1) * bin_w, 400 - int(hist[i] * 400 / 255)),
(100, 223, 255), -1)
cv.imshow('histImg', histImg)
cv.namedWindow('srcImg', cv.WINDOW_AUTOSIZE)
cv.createTrackbar('HueVal', 'srcImg', 0, 180, on_BinChg)
cv.imshow('srcImg', srcImg)
cv.waitKey(0)
cv.destroyAllWindows()
def generate_boxes(self):
record = set()
image = self.array
gray0 = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
timg = cv2.medianBlur(image, 9)
for c in range(3):
ch = [c, 0]
cv2.mixChannels([timg], [gray0], ch)
for l in range(self.N):
if l == 0:
gray = cv2.Canny(gray0, self.canny1, self.canny2,
self.canny3 * 2 + 1)
gray = cv2.dilate(
gray,
cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3)))
else:
_, gray = cv2.threshold(gray0, (l + 1) * 255 / self.N, 255,
cv2.THRESH_BINARY)
# gray=cv2.convertScaleAbs(gray)
# cv2.Mat()
_, contour, _ = cv2.findContours(gray, cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE)
for i in range(len(contour)):
approx = cv2.approxPolyDP(
contour[i],
cv2.arcLength(contour[i], True) * self.approx, True)
if len(approx) == 4 \
and 30000 < abs(cv2.contourArea(approx)) < self.image.size[0] * self.image.size[1] * 0.95 \
and cv2.isContourConvex(approx):
maxCosine = 0.0
for j in range(2, 5):
cosine = abs(
self.angle(approx[j % 4], approx[j - 2],
approx[j - 1]))
maxCosine = max(maxCosine, cosine)
if maxCosine < 0.3:
box = tuple((it[0][0], it[0][1]) for it in approx)
sorted_box = tuple(sorted(box))
if not self.sorted_box_in_record(
sorted_box, record):
record.add(sorted_box)
yield box
def got_frame(self, cvimg):
rgbimg = np.empty_like(cvimg)
cv2.mixChannels([cvimg], [rgbimg], [0, 2, 1, 1, 2, 0])
tmpimg = PImage.fromarray(rgbimg)
tmpimg.thumbnail((300, 300), PImage.ANTIALIAS)
tkimage = ImageTk.PhotoImage(tmpimg)
self.img_label.config({'image': tkimage})
self.img_label.photo = tkimage # prevent garbage collection
# Resize the image for display (or maybe should resize each output image)
if self.vid_writer:
self.vid_writer.write(rgbimg)
for tkfilter in self.filter_list:
tkfilter.got_frame(rgbimg)
def got_frame(self, cvimg):
rgbimg = np.empty_like(cvimg)
cv2.mixChannels([cvimg], [rgbimg], [0,2, 1,1, 2,0])
tmpimg = PImage.fromarray(rgbimg)
tmpimg.thumbnail((300, 300), PImage.ANTIALIAS)
tkimage = ImageTk.PhotoImage(tmpimg)
self.img_label.config({'image': tkimage})
self.img_label.photo = tkimage # prevent garbage collection
# Resize the image for display (or maybe should resize each output image)
if self.vid_writer:
self.vid_writer.write(rgbimg)
for tkfilter in self.filter_list:
tkfilter.got_frame(rgbimg)
def find_thresh(comp_img, target, bins):
## [Transform it to HSV]
hsv = cv.cvtColor(comp_img, cv.COLOR_BGR2HSV)
## [Transform it to HSV]
## [Use only the Hue value]
ch = (0, 0)
hue = np.empty(hsv.shape, hsv.dtype)
cv.mixChannels([hsv], [hue], ch)
bj = Hist_and_Backproj(bins, target, hue)
cv.imshow('bj', bj)
s = []
for x in bj:
s = list(set(s + list(set(x))))
print(s)
ret, thresh = cv.threshold(bj, 250, 255, cv.THRESH_BINARY)
return thresh
def __init__(self,iplimage):
# Rough-n-ready but it works dammit
# alpha = cv.CreateMat(iplimage.height,iplimage.width, cv.CV_8UC1)
alpha = np.zeros((iplimage.shape[0],iplimage.shape[1]), np.uint8)
# Zieht ein schwarzes Rechteck ueber das Bild
cv2.rectangle(alpha, (0, 0), (iplimage.shape[1],iplimage.shape[0]),
cv.ScalarAll(255) ,-1)
rgba = np.zeros((iplimage.shape[0], iplimage.shape[1], 4), np.uint8)
#cv2.Set(rgba, (1, 2, 3, 4))
cv2.mixChannels([iplimage, alpha],[rgba], [
0, 0, # rgba[0] -> bgr[2]
1, 1, # rgba[1] -> bgr[1]
2, 2, # rgba[2] -> bgr[0]
3, 3 # rgba[3] -> alpha[0]
])
self.__imagedata = rgba.tostring()
super(IplQImage,self).__init__(self.__imagedata, iplimage.shape[1],
iplimage.shape[0],
QtGui.QImage.Format_RGB32)
def apply_hist_mask(self, src, hist):
"""
Parameters
----------
src: np.array. Image frame
hist: array-like. Contains Hue probability distribuiton
Returns
-------
Returns image mask
"""
ch = (0, 0)
hsv = cv.cvtColor(src, cv.COLOR_BGR2HSV)
hue = np.empty(hsv.shape, hsv.dtype)
cv.mixChannels([hsv], [hue], ch)
backproj = cv.calcBackProject([hue], [0],
self.hist,
self.ranges,
scale=1)
return backproj
def Go_31(self, src, device='cpu'):
if device == 'cpu':
# CPU version.
dst = cv2.mixChannels([src], [self.color_cpu, self.alpha_cpu],
(0, 0, 1, 1, 2, 2, 3, 3))
elif device == 'numpy':
# NumPy version.
dst = [src[..., 0:3], src[..., 3]]
else:
# GPU version.
# print("GPU version not support mixChannels.")
dst = [self.color_gpu, self.alpha_gpu]
return dst
def __init__(self, drawingOverlay, last_gframe):
# convert red pixels to yellow for the overlay
self.drawingOverlay = np.zeros_like(drawingOverlay)
cv2.mixChannels([drawingOverlay], [self.drawingOverlay],
(0, 0, 2, 1, 2, 2))
self.shape = (drawingOverlay.shape[1], drawingOverlay.shape[0])
# Grab all pixels of overlay
xs, ys, zs = np.where(drawingOverlay > 0)
self.overlayPts = zip(xs, ys)
# Grab bounding rect
ptsMat = np.float32(map(lambda x: [x], self.overlayPts))
y, x, h, w = cv2.boundingRect(ptsMat)
x1, y1 = x, y
x2, y2 = x + w, y + h
# Create dense optical flow tracker
self.first_flow_pts = np.float32([[x1, y1], [x2, y1], [x2, y2],
[x1, y2]])
self.flow_tracker = OpticalFlowHomographyTracker(
last_gframe, self.first_flow_pts)
def Go31(self, src_color, src_alpha, device='cpu'):
if device == 'cpu':
# CPU version.
dst = cv2.mixChannels([src_color, src_alpha], [self.merged_cpu],
(0, 0, 1, 1, 2, 2, 3, 3))[0]
elif device == 'numpy':
# NumPy version.
self.merged_cpu[..., 0:3] = src_color
self.merged_cpu[..., 3] = src_alpha
dst = self.merged_cpu
else:
# GPU version.
# print("GPU version not support mixChannels.")
dst = self.merged_gpu
return dst
import cv2 as cv
import numpy as np
src = cv.imread("../../CPP_OpenCV/img/Lena.jpg")
dst = np.zeros(src.shape, dtype=np.uint8)
mv = cv.split(src)
mv[0][:, :] = 0
mv[1][:, :] = 0
mv[2][:, :] = 0
merged = cv.merge(mv)
cv.imshow("merged", merged)
cv.mixChannels([src], [dst], [0, 0, 1, 1, 2, 2])
cv.imshow("mixChannel", dst)
cv.waitKey(0)
def __getitem__(self, index_and_mode):
if self.disentangled:
index, sampling_mode = index_and_mode
else:
index = index_and_mode
mat_attr = self.mat_attrs[index]
#mat=self.mats[index]
#geom=self.geoms[index]
#illum=self.illums[index]
##### OpenCV version
# read image
image_rgb = cv2.cvtColor(
cv2.imread(
os.path.join(self.root, "renderings", self.files[index]), 1),
cv2.COLOR_BGR2RGB)
size = image_rgb.shape[0]
# read normals
normals_bgra = cv2.imread(
os.path.join(self.root, "normals", self.files[index][:-3] + "png"),
-1)
if (type(normals_bgra) is np.ndarray):
# if the normals exist, resize them and trasnform to RGB (is BGR when reading)
if normals_bgra.shape[0] != size:
normals_bgra = cv2.resize(normals_bgra, (size, size))
normals = np.ndarray((size, size, 4), dtype=np.uint8)
cv2.mixChannels([normals_bgra], [normals],
[0, 2, 1, 1, 2, 0, 3, 3])
else:
# otherwise, put the normals and a full mask
mask = np.ones((size, size, 1), np.uint8) * 255
normals = np.ndarray((size, size, 4), dtype=np.uint8)
cv2.mixChannels([image_rgb, mask], [normals],
[0, 0, 1, 1, 2, 2, 3, 3])
if self.mode == "test":
#slighlty erode mask so that the results are nicer
element = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
normals[:, :, 3] = cv2.dilate(normals[:, :, 3], element)
# add mask to image
image = np.ndarray(normals.shape, dtype=np.uint8)
cv2.mixChannels([image_rgb, normals], [image],
[0, 0, 1, 1, 2, 2, 6, 3])
if (self.use_illum):
illum = cv2.imread(
os.path.join(self.root, "illum", self.files[index]), -1)
if (not type(illum) is np.ndarray):
illum = extract_highlights(image)
illum = np.concatenate([illum, illum, illum],
axis=2) #3channels?
else:
if illum.ndim == 3: #RGB image
illum = cv2.cvtColor(
illum, cv2.COLOR_BGR2RGB) #or cv2.COLOR_BGR2GRAY
else:
illum = illum[:, :, np.newaxis] #image is already B&W
illum = np.concatenate([illum, illum, illum], axis=2)
else:
illum = torch.Tensor()
##### PIL version: faster but apply the alpha channel when resizing
#image = Image.open(os.path.join(self.root, "renderings", self.files[index]))
# try:
# normals = Image.open(os.path.join(self.root, "normals", self.files[index][:-3]+"png"))
# mask=get_alpha_channel(normals)
# except FileNotFoundError:
# #put the original image in place of the normals + full mask
# normals=image
# mask = Image.new('L',normals.size,255)
# normals.putalpha(mask)
# image.putalpha(mask)
# apply the transforms
if self.transform is not None:
if self.use_illum:
image, normals, illum = self.transform(image, normals, illum)
else:
image, normals = self.transform(image, normals)
# mask the normals
normals = normals * normals[3:]
# mask the input image if asked
if self.mask_input_bg:
image = image * image[3:]
if self.use_illum: illum = illum * image[3:]
if self.disentangled:
return image, normals, illum, torch.FloatTensor(
mat_attr), sampling_mode
else:
return image, normals, self.files[index][:-4].split(
"/")[-1], torch.FloatTensor(mat_attr)
import cv2 as cv
src = cv.imread("./test.png")
cv.namedWindow("input", cv.WINDOW_AUTOSIZE)
cv.imshow("input", src)
# 蓝色通道为零
mv = cv.split(src)
mv[0][:, :] = 0
dst1 = cv.merge(mv)
cv.imshow("output1", dst1)
# 绿色通道为零
mv = cv.split(src)
mv[1][:, :] = 0
dst2 = cv.merge(mv)
cv.imshow("output2", dst2)
# 红色通道为零
mv = cv.split(src)
mv[2][:, :] = 0
dst3 = cv.merge(mv)
cv.imshow("output3", dst3)
cv.mixChannels(src, dst3, [2, 0])
cv.imshow("output4", dst3)
cv.waitKey(0)
cv.destroyAllWindows()
mv = cv.split(src)
mv[0][:, :] = 0
dst1 = cv.merge(mv)
cv.imshow("output1", dst1)
# 绿色通道为零
mv = cv.split(src)
mv[1][:, :] = 0
dst2 = cv.merge(mv)
cv.imshow("output2", dst2)
# 红色通道为零
mv = cv.split(src)
mv[2][:, :] = 0
dst3 = cv.merge(mv)
cv.imshow("output3", dst3)
# cv.mixChannels(src, dst3, [2, 0])
# cv.imshow("output4", dst3)
dst = np.zeros(src.shape, dtype=np.uint8)
print(src.shape)
print(dst.shape)
cv.mixChannels([src], [dst], fromTo=[2, 0, 1, 1, 0, 2]) # 就是交换第一和第三通道
cv.imshow("output4", dst)
cv.waitKey(0)
cv.destroyAllWindows()
"""
mixChannels就是通道的交换与提取，【2,0】将输入矩阵的第三个通道数据复制到输出矩阵的第一个通道。
"""
def find_squares(img):
blurred = cv2.medianBlur(img, 9)
height, width, depth = blurred.shape
gray0 = blurred.copy()
# gray0 = cv2.convertScaleAbs(cv2.cvtColor(np.empty((height, width, 1), dtype=np.uint16), cv2.COLOR_GRAY2BGR))
# gray = np.zeros((blurred.size()[0], blurred.size()[1], 1), dtype=np.uint8)
squares = []
# find squares in every color plane of the image
for c in range(0, 3):
ch = [c, 0]
cv2.mixChannels(blurred, gray0, ch)
# try several threshold levels
threshold_level = 8
for l in range(0, threshold_level):
# Use Canny instead of zero threshold level!
# Canny helps to catch squares with gradient shading
if l == 0:
gray = cv2.Canny(gray0, 10, 20, apertureSize=3)
# Dilate helps to remove potential holes between edge segments
gray = cv2.dilate(gray, np.ones((11, 11), "uint8"))
else:
# gray = 1 if (gray0.any() >= (l+1) * 255 / threshold_level) else 0
gray = cv2.cvtColor(
cv2.threshold(gray0, int(l / float(threshold_level) * 255), 128, cv2.THRESH_BINARY_INV)[1],
cv2.COLOR_BGR2GRAY,
)
# Find contours and store them in a list
# print gray
contours = cv2.findContours(gray, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[1]
# Test contours
for i in range(0, len(contours)):
# approximate contour with accuracy proportional
# to the contour perimeter
arclength = cv2.arcLength(np.array(contours[i], copy=True), True) * 0.02
approx = cv2.approxPolyDP(contours[i], arclength, True)
# Note: absolute value of an area is used because
# area may be positive or negative - in accordance with the
# contour orientation
if (
len(approx) == 4
and abs(cv2.contourArea(np.array(approx, copy=True))) > 1000
and cv2.isContourConvex(np.array(approx, copy=True))
):
maxCosine = 0
for j in range(2, 5):
cosine = abs(angle(approx[j % 4], approx[j - 2], approx[j - 1]))
maxCosine = max([maxCosine, cosine])
if maxCosine > math.pi:
squares.append(approx)
return squares
args = parser.parse_args()
src = cv.imread(args.input)
if src is None:
print('Could not open or find the image:', args.input)
exit(0)
## [Read the image]
## [Transform it to HSV]
hsv = cv.cvtColor(src, cv.COLOR_BGR2HSV)
## [Transform it to HSV]
## [Use only the Hue value]
ch = (0, 0)
hue = np.empty(hsv.shape, hsv.dtype)
cv.mixChannels([hsv], [hue], ch)
## [Use only the Hue value]
## [Create Trackbar to enter the number of bins]
window_image = 'Source image'
cv.namedWindow(window_image)
bins = 25
cv.createTrackbar('* Hue bins: ', window_image, bins, 180, Hist_and_Backproj )
Hist_and_Backproj(bins)
## [Create Trackbar to enter the number of bins]
## [Show the image]
cv.imshow(window_image, src)
cv.waitKey()
## [Show the image]
# image0[:, :, 0] = 255
# image0[:, :, 1] = 255
# image0[:, :, 2] = 255
cv2.namedWindow('image0', cv2.WINDOW_AUTOSIZE)
cv2.imshow('image0', image0)
# # 蓝色通道为零
mv = cv2.split(image0)
mv[0][:, :] = 0
#mv[2][:, :] = 0
dst1 = cv2.merge(mv)
cv2.imshow('dst1', dst1)
# # 绿色通道为零
mv = cv2.split(image0)
mv[1][:, :] = 0
dst2 = cv2.merge(mv)
cv2.imshow('dst2', dst2)
# # 红色通道为零
mv = cv2.split(image0)
mv[2][:, :] = 0
dst3 = cv2.merge(mv)
cv2.imshow('dst3', dst3)
cv2.mixChannels([dst1], [dst3], [0, 0]) #一定注意输入和输出外层的括号
cv2.imshow('image1', dst3)
cv2.waitKey(0)
cv2.destroyAllWindows()
import numpy as np
src = cv.imread("../data/images/cos.jpg")
cv.imshow("source image", src)
## 通道分离
b, g, r = cv.split(src)
cv.imshow("blue", b)
cv.imshow("green", g)
cv.imshow("red", r)
## 通道合并，将一个通道所有值设置为0再合并
for i in range(3):
bgr = cv.split(src)
bgr[i][:, :] = 0 # 将第i个通道设置为0
dst = cv.merge(bgr)
cv.imshow("merge channel without {}".format(i), dst)
## mixChannels 将输入数组的指定通道复制到输出数组的指定通道
# 需要初始化输出矩阵的大小
dst = np.zeros_like(src)
# 两个两个一组，0，1；1，2；2，0，src的第一个通道放到dst第二个换，src的第二个通道与放到dst第三个，src的第三个通道与放到dst第一个通道
from_to = [0, 1, 1, 2, 2, 0]
# 加上[]，将src和dst变成列表，否则结果不正常，参考：https://stackoverflow.com/questions/42329901/opencv-python-cv2-mixchannels
cv.mixChannels([src], [dst], from_to)
cv.imshow("mixChannels", dst)
cv.waitKey(0)
cv.destroyAllWindows()
def main():
Posiciones = [] # Guardo las posiciones iniciales de configuracion
featuringRead = np.loadtxt('featuring.out', delimiter=' ')
featuringRead = np.float32(featuringRead)
# Apply KMeans
compactness, labels, centers = cv2.kmeans(
featuringRead,
K=3,
bestLabels=None,
criteria=(cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_MAX_ITER, 20, 0),
attempts=1,
flags=cv2.KMEANS_RANDOM_CENTERS)
A = featuringRead[labels.ravel() == 0]
B = featuringRead[labels.ravel() == 1]
C = featuringRead[labels.ravel() == 2]
MIN_OBJECT_AREA = 20 * 20
MAX_OBJECT_AREA = 40 * 40
Multi = cv2.imread("multi700x490.jpg", 1)
Multihsv = cv2.cvtColor(Multi, cv2.COLOR_BGR2HSV)
Multihue = np.zeros(Multihsv.shape, dtype=np.uint8)
cv2.mixChannels([Multihsv], [Multihue], [0, 0])
rospy.init_node("Analisis_de_Color")
"""reset_cameras()
close_camera("left")
close_camera("right")
close_camera("head")"""
open_camera("right", 960, 600)
subscribe_to_camera("right")
#screen_pub = rospy.Publisher('/robot/xdisplay', Image, queue_size=10)
endpoint_sub = rospy.Subscriber('/robot/limb/right/endpoint_state',
EndpointState,
callback=endpoint_callback)
state_sub = rospy.Subscriber('robot/navigators/right_navigator/state',
NavigatorState,
callback=on_state)
#rate = rospy.Rate(3)
print 'Ahora graba posiciones'
# Posiciones
########3########
#################
#2######0#######1
#################
########4########
#
while 1:
if button0:
Posiciones.append([x_ini, y_ini])
while button0:
continue
if button1 or button2:
break
print 'posiciones grabadas'
blank_image = np.zeros((600, 960, 3), np.uint8)
for i in range(20):
cv2.line(blank_image, (int((960 / 20) * (i + 1)), 0), (int(
(960 / 20) * (i + 1)), 600), (255, 0, 0), 5)
cv2.line(blank_image, (0, int(600 / 20) * (i + 1)),
(960, int((600 / 20) * (i + 1))), (255, 0, 0), 5)
cv2.circle(blank_image, (int(centers[0][0]), int(centers[0][1])), 20,
(0, 255, 0), -1)
cv2.circle(blank_image, (int(centers[1][0]), int(centers[1][1])), 20,
(0, 255, 0), -1)
cv2.circle(blank_image, (int(centers[2][0]), int(centers[2][1])), 20,
(0, 255, 0), -1)
for i in range(len(A)):
cv2.circle(blank_image, (int(A[i][0]), int(A[i][1])), 4, (0, 255, 255),
-1)
for i in range(len(B)):
cv2.circle(blank_image, (int(B[i][0]), int(B[i][1])), 4, (0, 255, 255),
-1)
for i in range(len(C)):
cv2.circle(blank_image, (int(C[i][0]), int(C[i][1])), 4, (0, 255, 255),
-1)
Muevee = Mueve('right', Posiciones)
erodeElement = cv2.getStructuringElement(
cv2.MORPH_RECT, (3, 3)) # selecciono el tipo de kernel
while 1:
Etiquetas = [] # Guardo las etiquetas de KMeans
Angulo = []
Dimensiones = []
Coordenadas = []
SetFrames = []
Grays = []
if img.getImg() is None:
continue
while len(SetFrames) < 20:
SetFrames.append(np.copy(img.getImg()))
Grays.append(0)
Color = img.getImg()
gray = cv2.cvtColor(Color, cv2.COLOR_BGR2GRAY)
gray = cv2.equalizeHist(gray)
gray = cv2.GaussianBlur(gray, (5, 5), 0, 0)
"""sobel64f = cv2.Sobel(gray,cv2.CV_64F,1,1,ksize=5)
abs_sobel64f = np.absolute(sobel64f)
sobel_8u = np.uint8(abs_sobel64f)
sobel_8u = cv2.erode(sobel_8u,erodeElement)"""
gray = cv2.Canny(gray, 100, 150, 3)
"""for m in range(gray.shape[:2][0]):
for n in range(gray.shape[:2][1]):
if gray[m][n]==0:
gray[m][n]=sobel_8u[m][n]"""
for i in range(len(SetFrames)):
Grays[i] = cv2.cvtColor(SetFrames[i], cv2.COLOR_BGR2GRAY)
Grays[i] = cv2.equalizeHist(Grays[i])
Grays[i] = cv2.GaussianBlur(Grays[i], (5, 5), 0, 0)
Grays[i] = cv2.Canny(Grays[i], 100, 150, 3)
for i in range(len(Grays)):
gray = cv2.bitwise_or(gray, Grays[i])
cv2.imshow("gray", gray)
(contours, hierarchy) = cv2.findContours(gray.copy(),
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_NONE)
if len(hierarchy) > 0:
index = 0
while index != -1:
moment = cv2.moments(contours[index])
area = moment['m00']
if area > MIN_OBJECT_AREA and area < MAX_OBJECT_AREA:
rect = cv2.minAreaRect(contours[index])
roi = crop_minAreaRect(Color, rect)
if roi.shape[:2][1] > 0 and roi.shape[:2][0] > 0:
hsv = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV)
hue = np.zeros(hsv.shape, dtype=np.uint8)
cv2.mixChannels([hsv], [hue], [0, 0])
hist = cv2.calcHist([hue], [0], None, [180], [0, 180])
cv2.normalize(hist, hist, 0, 255, cv2.NORM_MINMAX)
backproj = cv2.calcBackProject([Multihue], [0], hist,
[0, 180], 1)
ret, backproj = cv2.threshold(backproj, 127, 255,
cv2.THRESH_BINARY)
receiver = CentroMasa(backproj, Color)
dist = [
math_calc_dist(receiver, centers[0]),
math_calc_dist(receiver, centers[1]),
math_calc_dist(receiver, centers[2])
]
min_index, min_value = min(enumerate(dist),
key=operator.itemgetter(1))
#print 'centro: ',centers[min_index],' distancia: ',dist[min_index]
Coordenadas.append(
list([
int(moment['m10'] / area),
int(moment['m01'] / area)
]))
Etiquetas.append(min_index)
Angulo.append(rect[2])
Dimensiones.append(rect[1])
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
cv2.circle(blank_image,
(int(receiver[0]), int(receiver[1])), 5,
(0, 0, 255), -1)
cv2.drawContours(Color, [box], 0, (0, 0, 255), 2)
index = hierarchy[0][index][0]
cv2.imshow("Imagen Filtrada", Color)
cv2.imshow("Enrejado", blank_image)
#msgsub = cv_bridge.CvBridge().cv2_to_imgmsg(blank_image, encoding="8UC3")
#screen_pub.publish(msgsub)
Color1 = []
Color2 = []
Color3 = []
print Etiquetas
for i in range(len(Coordenadas)):
if Etiquetas[i] == 0:
Color1.append(i)
elif Etiquetas[i] == 1:
Color2.append(i)
else:
Color3.append(i)
for i in Color1:
Muevee.CA(Coordenadas[i][0], Coordenadas[i][1], Etiquetas[i],
Angulo[i], Dimensiones[i])
for i in Color2:
Muevee.CA(Coordenadas[i][0], Coordenadas[i][1], Etiquetas[i],
Angulo[i], Dimensiones[i])
for i in Color3:
Muevee.CA(Coordenadas[i][0], Coordenadas[i][1], Etiquetas[i],
Angulo[i], Dimensiones[i])
Muevee.randomm()
Muevee.mover_baxter('base', Muevee.pose[:3], Muevee.pose[3:6])
if cv2.waitKey(1) & 0xFF == ord(
'q'): # Indicamos que al pulsar "q" el programa se cierre
break
#rate.sleep()
cv2.destroyAllWindows()
import cv2
import numpy as np
if __name__ == '__main__':
roi = cv2.imread('hope.jpg')
hsv = cv2.cvtColor(roi,cv2.COLOR_BGR2HSV)
target = cv2.imread('img3.jpg')
hsvt = cv2.cvtColor(target,cv2.COLOR_BGR2HSV)
#mix Channels
hue = np.empty(hsv.shape, np.float32)
cv2.mixChannels(hsv.astype('float32'),hue, [0, 0])
# calculating object histogram
roihist = cv2.calcHist([hue],[0, 0], None, [180, 256], [0, 180, 0, 256] )
# normalize histogram and apply backprojection
cv2.normalize(roihist,roihist,0,255,cv2.NORM_MINMAX)
dst = cv2.calcBackProject([hsvt],[0,0],roihist,[0,180,0,256],1)
#a minsziftem
term_crit = ( cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1 )
ret, track_window = cv2.meanShift(dst, track_window, term_crit)
# Draw it on image
x,y,w,h = track_window
img2 = cv2.rectangle(frame, (x,y), (x+w,y+h), 255,2)
cv2.imshow('img2',img2)
# Now convolute with circular disc
src = 'cartoon.jpg'
input_image = cv.imread(src)
if input_image is None:
print('Could not load image: ', input_image)
exit(0)
# Splitting image into RGB channels:
blue, green, red = cv.split(input_image)
print(blue.shape)
# We create a dummy 3D array
blue_channel = np.zeros(input_image.shape, input_image.dtype)
green_channel = np.zeros(input_image.shape, input_image.dtype)
red_channel = np.zeros(input_image.shape, input_image.dtype)
# We match each color channel to a 3D dimension:
# Blue Rendering : [blue; 0; 0]
# Green Rendering: [0; green; 0]
# Red Rendering: [0; 0; red]
cv.mixChannels([blue, green, red], [blue_channel], [0, 0])
cv.mixChannels([blue, green, red], [green_channel], [1, 1])
cv.mixChannels([blue, green, red], [red_channel], [2, 2])
cv.imshow('Blue Channel', blue_channel)
cv.imshow('Green Channel', green_channel)
cv.imshow('Red Channel', red_channel)
cv.waitKey(0)
cv.destroyAllWindows()
def get_feed(self):
global a_time_to_die, feed_data, feed_width, feed_height, feed_ready
scanner = zbar.ImageScanner()
scanner.parse_config('enable')
shift = 1
while 1:
if a_time_to_die:
return
try:
HOST = '' # The remote host
PORT = 12345 # The same port as used by the server
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect((HOST, PORT))
s.send('Yo!')
data = data = recvall(s)
s.close()
data = pickle.loads(data)
#print 'Received img'
height, width, depth = data.shape
if effect == smiley_face:
try:
data = cv2.flip(detect_faces(cv2.flip(data, 0)),0)
except:
print "face fail"
elif effect == target_face:
try:
data = cv2.flip(detect_faces_2(cv2.flip(data, 0)),0)
except:
print "face 2 fail"
elif effect == trippy_colours:
try:
shift += 1
data += shift % 256
except:
print "shift fail"
elif effect == trippy_colours_2:
try:
shift += 1
for x in xrange(shift % 3):
data2 = cv2.copy(data)
cv2.mixChannels(data2,data, [(0,1),(1,2),(2,1)])
except:
print "mario fail"
elif effect == vertical_hold:
try:
shift += 10
M = numpy.float32([[1,0,0],[0,1,shift%height]])
data1 = cv2.warpAffine(data,M,(width,height))
M = numpy.float32([[1,0,0],[0,1,(shift%height)-height]])
data2 = cv2.warpAffine(data,M,(width,height))
data = data1 + data2
M = numpy.float32([[1,0,shift%width],[0,1,0]])
data1 = cv2.warpAffine(data,M,(width,height))
M = numpy.float32([[1,0,(shift%width)-width],[0,1,0]])
data2 = cv2.warpAffine(data,M,(width,height))
data = data1 + data2
except:
print "vhold fail"
elif effect == colour_band:
try:
shift += 5
b = height / 20
top = shift % (width - b)
# band of colour
data_g = cv2.cvtColor(cv2.cvtColor(data, cv2.COLOR_BGR2GRAY), cv2.CV_GRAY2BGR)
data_g[0:width, top:top+b] = data[0:width, top:top+b]
data = data_g
# inversion
data_i = 255 - data
data[0:width, top:top+b] = data_i[0:width, top:top+b]
except:
print "vhold fail"
data = cv2.cvtColor(data,cv2.COLOR_BGR2RGB)
feed_data = data.tostring()
feed_width = width
feed_height = height
feed_ready = True
sleep(.001)
except:
s.close()
print("fail")
sleep(25)
src = cv.imread("../images/1.png")
cv.namedWindow("input", cv.WINDOW_AUTOSIZE)
cv.imshow("input", src)
# 蓝色通道为零
mv = cv.split(src)
mv[0][:, :] = 0
dst1 = cv.merge(mv)
cv.imshow("output1", dst1)
# 绿色通道为零
mv = cv.split(src)
mv[1][:, :] = 0
dst2 = cv.merge(mv)
cv.imshow("output2", dst2)
# 红色通道为零
mv = cv.split(src)
mv[2][:, :] = 0
dst3 = cv.merge(mv)
cv.imshow("output3", dst3)
dst = np.zeros(src.shape, dtype=np.uint8)
print(src.shape)
print(dst.shape)
cv.mixChannels([src], [dst], fromTo=[2, 0, 1, 1, 0, 2])
cv.imshow("output4", dst)
cv.waitKey(0)
cv.destroyAllWindows()
#
#
#
import cv2
src = cv2.imread("d://pics//212121.jpg")
windowImage = 'inputImage'
cv2.namedWindow(windowImage, cv2.WINDOW_NORMAL)
cv2.namedWindow('backProject', cv2.WINDOW_NORMAL)
# cv2.namedWindow('histogram', cv2.WINDOW_NORMAL);
hsv = cv2.cvtColor(src, cv2.COLOR_BGR2HSV)
nChannels = [0, 0]
cv2.mixChannels(hsv, hsv, nChannels)
hue = hsv
bins = 12
cv2.imshow(windowImage, src)
histBase = cv2.calcHist([hsv], [0], None, [bins], [0, 180])
cv2.normalize(hsv, hue, 0, 256, cv2.NORM_MINMAX, -1, None)
hue = cv2.calcBackProject([hue], [0], 0, [0, 180], 1)
# def histAndBackProjection(pos , tmp):
# cv2.calcHist(hue,[1],None,pos,[0,180]);
# cv2.normalize(hue,hue,0,1,cv2.NORM_MINMAX,-1,None);
#
# hue = cv2.calcBackProject([hue],1,0,[0,180],(0,255,255));
#
def find_squares(img):
blurred = cv2.medianBlur(img, 9)
height, width, depth = blurred.shape
gray0 = blurred.copy()
#gray0 = cv2.convertScaleAbs(cv2.cvtColor(np.empty((height, width, 1), dtype=np.uint16), cv2.COLOR_GRAY2BGR))
# gray = np.zeros((blurred.size()[0], blurred.size()[1], 1), dtype=np.uint8)
squares = []
# find squares in every color plane of the image
for c in range(0, 3):
ch = [c, 0]
cv2.mixChannels(blurred, gray0, ch)
# try several threshold levels
threshold_level = 8
for l in range(0, threshold_level):
# Use Canny instead of zero threshold level!
# Canny helps to catch squares with gradient shading
if l == 0:
gray = cv2.Canny(gray0, 10, 20, apertureSize=3)
# Dilate helps to remove potential holes between edge segments
gray = cv2.dilate(gray, np.ones((11, 11), 'uint8'))
else:
#gray = 1 if (gray0.any() >= (l+1) * 255 / threshold_level) else 0
gray = cv2.cvtColor(
cv2.threshold(gray0, int(l / float(threshold_level) * 255),
128, cv2.THRESH_BINARY_INV)[1],
cv2.COLOR_BGR2GRAY)
# Find contours and store them in a list
#print gray
contours = cv2.findContours(gray, cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE)
contours = contours[1]
# Test contours
for i in range(0, len(contours)):
# approximate contour with accuracy proportional
# to the contour perimeter
arclength = cv2.arcLength(np.array(contours[i], copy=True),
True) * 0.02
approx = cv2.approxPolyDP(contours[i], arclength, True)
# Note: absolute value of an area is used because
# area may be positive or negative - in accordance with the
# contour orientation
if len(approx) == 4 and abs(
cv2.contourArea(np.array(approx, copy=True))
) > 1000 and cv2.isContourConvex(np.array(approx, copy=True)):
maxCosine = 0
for j in range(2, 5):
cosine = abs(
angle(approx[j % 4], approx[j - 2], approx[j - 1]))
maxCosine = max([maxCosine, cosine])
if maxCosine > math.pi:
squares.append(approx)
return squares
