def findPairs(surf, objectKeypoints, objectDescriptors, imageKeypoints, imageDescriptors):
ptpairs = []
N = len(objectKeypoints)
K = 2 # K as in K nearest neighbors
dim = surf.descriptorSize()
m_object = cv.asMat(objectDescriptors).reshape(1, N)
m_image = cv.asMat(imageDescriptors).reshape(1, len(imageKeypoints))
flann = cv.Index(m_image,cv.KDTreeIndexParams(4))
indices = cv.Mat(N, K, cv.CV_32S)
dists = cv.Mat(N, K, cv.CV_32F)
flann.knnSearch(m_object, indices, dists, K, cv.SearchParams(250))
indices = indices[:,0].ravel()
dists = dists.ndarray
for i in xrange(N):
if dists[i,0] < 0.6*dists[i,1]:
ptpairs.append((objectKeypoints[i].pt, imageKeypoints[int(indices[i])].pt))
return ptpairs
def findPairs(surf, objectKeypoints, objectDescriptors, imageKeypoints,
imageDescriptors):
ptpairs = []
N = len(objectKeypoints)
K = 2 # K as in K nearest neighbors
dim = surf.descriptorSize()
m_object = cv.asMat(objectDescriptors).reshape(1, N)
m_image = cv.asMat(imageDescriptors).reshape(1, len(imageKeypoints))
flann = cv.Index(m_image, cv.KDTreeIndexParams(4))
indices = cv.Mat(N, K, cv.CV_32S)
dists = cv.Mat(N, K, cv.CV_32F)
flann.knnSearch(m_object, indices, dists, K, cv.SearchParams(250))
indices = indices[:, 0].ravel()
dists = dists.ndarray
for i in xrange(N):
if dists[i, 0] < 0.6 * dists[i, 1]:
ptpairs.append(
(objectKeypoints[i].pt, imageKeypoints[int(indices[i])].pt))
return ptpairs
def get_polygons(contours, hierarchy,
convextest=False,
# hole=False,
nsides=5,
min_area=100,
perimeter_smooth_factor=0.001,
**kw):
'''
'''
polygons = []
brs = []
areas = []
for cont, hi in zip(contours, hierarchy.tolist()):
cont = cv.asMat(cont)
# for i in [0.01]:
m = cv.arcLength(cont, True)
result = cv.approxPolyDP_int(cont, m * perimeter_smooth_factor, True)
res_mat = cv.asMat(result)
area = abs(cv.contourArea(res_mat))
# print 'areas', area
# if hole:
# hole_flag = hi[3] != -1
# else:
# hole_flag = True
# if area > min_area:
# print 'area', area,
# print 'hole', hole_flag
# print 'hi', hi
# print 'sides', len(result),
# print 'convext', cv.isContourConvex(res_mat),
# ch = cv.asMat(cv.convexHull_int(cont))
# ch = cv.asMat(ch.ndarray.flatten())
# seq = cv.convexityDefects(cont, ch, cv.createMemStorage(0))
#
# if not hole_flag:
# continue
if not len(result) > nsides:
continue
if not area > min_area:
continue
if convextest and not cv.isContourConvex(res_mat):
continue
polygons.append(result)
brs.append(cv.boundingRect(res_mat))
areas.append(area)
return polygons, brs, areas
def mouse_call_back(event, x, y, flags, user_data):
global seed
# 右键松开时，初始化种子图像
if event == cv.CV_EVENT_RBUTTONUP:
img2[:] = img[:]
markers[:] = 0
seed = 1
cv.imshow("Watershed Demo", img2)
if seed == len(marks_color): return
# 左键按下时，在种子图像上添加种子
if flags == cv.CV_EVENT_FLAG_LBUTTON:
pt = cv.Point(x, y)
cv.circle(markers, pt, 5, cv.Scalar(seed, seed, seed, seed),
cv.CV_FILLED)
cv.circle(img2, pt, 5, marks_color[seed], cv.CV_FILLED)
cv.imshow("Watershed Demo", img2)
# 左键松开时，使用watershed进行图像分割
if event == cv.CV_EVENT_LBUTTONUP:
seed += 1
tmp_markers = markers.clone()
cv.watershed(img, tmp_markers)
color_map = tmp_markers[:].astype(np.int)
img3 = img2.clone()
img4 = cv.asMat(palette[color_map])
cv.addWeighted(img3, 1.0, img4, mask_opacity, 0, img3)
cv.imshow("Watershed Demo", img3)
def redraw(self):
M = cv.asMat(self.m, force_single_channel=True)
size = cv.Size(int(self.size[0,0]), int(self.size[0,1]))
img2 = cv.Mat()
if size.width > 0 and size.height > 0:
cv.warpAffine(self.img, img2, M, size, borderValue=cv.CV_RGB(255,255,255))
cv.imshow("Affine Demo", img2)
def mouse_call_back(event, x, y, flags, user_data):
global seed
# 右键松开时，初始化种子图像
if event == cv.CV_EVENT_RBUTTONUP:
img2[:] = img[:]
markers[:] = 0
seed = 1
cv.imshow("Watershed Demo", img2)
if seed == len(marks_color): return
# 左键按下时，在种子图像上添加种子
if flags == cv.CV_EVENT_FLAG_LBUTTON:
pt = cv.Point(x, y)
cv.circle(markers, pt, 5, cv.Scalar(seed,seed,seed,seed), cv.CV_FILLED)
cv.circle(img2, pt, 5, marks_color[seed], cv.CV_FILLED)
cv.imshow("Watershed Demo", img2)
# 左键松开时，使用watershed进行图像分割
if event == cv.CV_EVENT_LBUTTONUP:
seed += 1
tmp_markers = markers.clone()
cv.watershed(img, tmp_markers)
color_map = tmp_markers[:].astype(np.int)
img3 = img2.clone()
img4 = cv.asMat( palette[color_map] )
cv.addWeighted(img3, 1.0, img4, mask_opacity, 0, img3)
cv.imshow("Watershed Demo", img3)
def affine(self):
self.img2 = cv.Mat()
M = cv.asMat(self.m, force_single_channel=True)
cv.warpAffine(self.img1,
self.img2,
M,
self.img1.size(),
borderValue=cv.CV_RGB(255, 255, 255))
def locatePlanarObject(ptpairs, src_corners, dst_corners):
import numpy as _n
n = len(ptpairs)
if n < 4:
return 0
pt1 = cv.asMat(_n.array([cv.asndarray(pair[0]) for pair in ptpairs]))
pt2 = cv.asMat(_n.array([cv.asndarray(pair[1]) for pair in ptpairs]))
h = cv.findHomography(pt1, pt2, method=cv.RANSAC, ransacReprojThreshold=5)[:]
for i in range(4):
x = src_corners[i].x
y = src_corners[i].y
Z = 1./(h[2,0]*x + h[2,1]*y + h[2,2])
X = (h[0,0]*x + h[0,1]*y + h[0,2])*Z
Y = (h[1,0]*x + h[1,1]*y + h[1,2])*Z
dst_corners[i] = cv.Point(int(X), int(Y))
return 1
def redraw(self):
M = cv.asMat(self.m, force_single_channel=True)
size = cv.Size(int(self.size[0, 0]), int(self.size[0, 1]))
img2 = cv.Mat()
if size.width > 0 and size.height > 0:
cv.warpAffine(self.img,
img2,
M,
size,
borderValue=cv.CV_RGB(255, 255, 255))
cv.imshow("Affine Demo", img2)
def redraw(self):
img2 = cv.Mat()
element = cv.asMat(self.structing_element, force_single_channel=True)
if self.process_type.startswith("MORPH_"):
type = getattr(cv, self.process_type)
cv.morphologyEx(self.img, img2, type, element, iterations=self.iter)
else:
func = getattr(cv, self.process_type)
func(self.img, img2, element, iterations=self.iter)
cv.imshow("Morphology Demo", img2)
def redraw(self):
img2 = cv.Mat()
element = cv.asMat(self.structing_element, force_single_channel=True)
if self.process_type.startswith("MORPH_"):
type = getattr(cv, self.process_type)
cv.morphologyEx(self.img, img2, type, element, iterations=self.iter)
else:
func = getattr(cv, self.process_type)
func(self.img, img2, element, iterations=self.iter)
cv.imshow("Morphology Demo", img2)
def locatePlanarObject(ptpairs, src_corners, dst_corners):
import numpy as _n
n = len(ptpairs)
if n < 4:
return 0
pt1 = cv.asMat(_n.array([cv.asndarray(pair[0]) for pair in ptpairs]))
pt2 = cv.asMat(_n.array([cv.asndarray(pair[1]) for pair in ptpairs]))
h = cv.findHomography(pt1, pt2, method=cv.RANSAC,
ransacReprojThreshold=5)[:]
for i in range(4):
x = src_corners[i].x
y = src_corners[i].y
Z = 1. / (h[2, 0] * x + h[2, 1] * y + h[2, 2])
X = (h[0, 0] * x + h[0, 1] * y + h[0, 2]) * Z
Y = (h[1, 0] * x + h[1, 1] * y + h[1, 2]) * Z
dst_corners[i] = cv.Point(int(X), int(Y))
return 1
def get_focus_measure(src, kind):
# from numpy import r_
# from scipy import fft
# from pylab import plot, show
# w = 100
# h = 100
# x = (640 - w) / 2
# y = (480 - h) / 2
# src = crop(src, x, y, w, h)
# src = grayspace(src)
# d = src.ndarray
#
# # print d[0]
# # print d[-1]
# fftsig = fft(d)
# d = abs(fftsig)
# print d.shape
# dst = src.clone()
# cv.convertScaleAbs(cv.asMat(d), dst, 1, 0)
# return dst
# xs = xrange(len(ys))
# plot(xs, ys)
# show()
# N = len(d)
# f = 50000 * r_[0:(N / 2)] / N
# n = len(f)
# # print f
# d = d.transpose()
# d = abs(fftsig[:n]) / N
# print d
# # plot(f, d[0], 'b', f, d[1], 'g', f, d[2], 'r')
# plot(f, d)
# show()
planes = cv.vector_Mat()
src = cv.asMat(src)
laplace = cv.Mat(src.size(), cv.CV_16SC1)
colorlaplace = cv.Mat(src.size(), cv.CV_8UC3)
cv.split(src, planes)
for plane in planes:
cv.Laplacian(plane, laplace, 3)
cv.convertScaleAbs(laplace, plane, 1, 0)
cv.merge(planes, colorlaplace)
f = colorlaplace.ndarray.flatten()
# f.sort()
# print f[-int(len(f) * 0.1):], int(len(f) * 0.1), len(f)
# len(f)
return f[-int(len(f) * 0.1):].mean()
def process(self, input_images, connected_outs):
if len(input_images) == 0:
return FAIL
src = input_images['Input']
dist_res = int( self.getParamContent('Distance resolution') )
angle_res = int( self.getParamContent('Angle resolution (degrees)') )
acc_thresh = int( self.getParamContent('Accumulator threshold') )
min_length = int( self.getParamContent('Minimum length') )
max_gap = int( self.getParamContent('Maximum gap') )
choice = self.getParamContent("Type of Hough transform")
if src.ndim > 2:
print "In '%s': The hough transform takes a binary image (or 8-bit) as input." %self.name
return FAIL
color_dst = numpy.empty( (src.shape[0], src.shape[1], 3),dtype='uint8' )
pycv.cvtColor( pycv.asMat(src), pycv.asMat(color_dst), pycv.CV_GRAY2BGR )
if choice == "Standard":
lines = pycv.HoughLines( pycv.asMat(src), dist_res, pycv.CV_PI/angle_res, acc_thresh )
margin = 0.04
n=8
pi = math.pi
h,w = src.shape[0:2]
for i in range(min(len(lines), int(self.getParamContent("draw # lines")))):
l = lines[i]
rho = l[0]
theta = l[1]
if theta > 3*pi/4: theta-=pi
if abs(rho)<w/n and abs(theta)<margin: pass
elif abs(rho)>w-w/n and abs(theta)<margin: pass
elif abs(rho)<h/n and abs(theta-pi/2)<margin: pass
elif abs(rho)>h-h/n and abs(theta-pi/2)<margin: pass
else:
continue
a = math.cos(theta)
b = math.sin(theta)
x0 = a*rho
y0 = b*rho
pt1 = pycv.Point( int(round(x0 + 2000*(-b))), int(round(y0 + 2000*(a))) )
pt2 = pycv.Point( int(round(x0 - 2000*(-b))), int(round(y0 - 2000*(a))) )
pycv.line( pycv.asMat(color_dst), pt1, pt2, pycv.CV_RGB(random.randint(0,255),
random.randint(0,255),
random.randint(0,255)), 2, 8 )
else:
lines = pycv.HoughLinesP( pycv.asMat(src), dist_res,
pycv.CV_PI/angle_res, acc_thresh, min_length, max_gap )
for l in lines:
pycv.line( pycv.asMat(color_dst), pycv.Point(int(l[0]), int(l[1])),
pycv.Point(int(l[2]), int(l[3])),
pycv.CV_RGB(*getRandColor()), 2, 8 )
self.lines = [(item[0],item[1]) for item in lines]
return {self.output_names[0] : color_dst, self.output_names[1]:self.lines}
def __init__(self):
#读入图像
img = cv.imread("lena_full.jpg")
img2 = cv.Mat()
cv.cvtColor(img, img2, cv.CV_BGR2GRAY)
img = cv.Mat()
cv.resize(img2, img, cv.Size(N, N))
self.fimg = fft.fft2(img[:]) # 图像的频域信号
mag_img = np.log10(np.abs(self.fimg))
# 创建计算用图像
filtered_img = np.zeros((N, N), dtype=np.float)
self.mask = np.zeros((N, N), dtype=np.float)
self.mask_img = cv.asMat(self.mask) # 在self.mask上绘制多边形用的图像
# 创建数据源
self.data = ArrayPlotData(mag_img=fft.fftshift(mag_img),
filtered_img=filtered_img,
mask_img=self.mask)
# 创建三个图像绘制框以及容器
meg_plot, img = self.make_image_plot("mag_img")
mask_plot, _ = self.make_image_plot("mask_img")
filtered_plot, _ = self.make_image_plot("filtered_img")
self.plot = HPlotContainer(meg_plot, mask_plot, filtered_plot)
# 创建套索工具
lasso_selection = LassoSelection(component=img)
lasso_overlay = LassoOverlay(lasso_selection=lasso_selection,
component=img,
selection_alpha=0.3)
img.tools.append(lasso_selection)
img.overlays.append(lasso_overlay)
self.lasso_selection = lasso_selection
# 监听套索工具的事件、开启时钟事件
lasso_selection.on_trait_change(self.lasso_updated,
"disjoint_selections")
self.timer = Timer(50, self.on_timer)
def __init__(self):
#读入图像
img = cv.imread("lena_full.jpg")
img2 = cv.Mat()
cv.cvtColor(img, img2, cv.CV_BGR2GRAY)
img = cv.Mat()
cv.resize(img2, img, cv.Size(N, N))
self.fimg = fft.fft2(img[:]) # 图像的频域信号
mag_img = np.log10(np.abs(self.fimg))
# 创建计算用图像
filtered_img = np.zeros((N, N), dtype=np.float)
self.mask = np.zeros((N, N), dtype=np.float)
self.mask_img = cv.asMat(self.mask) # 在self.mask上绘制多边形用的图像
# 创建数据源
self.data = ArrayPlotData(
mag_img = fft.fftshift(mag_img),
filtered_img = filtered_img,
mask_img = self.mask
)
# 创建三个图像绘制框以及容器
meg_plot, img = self.make_image_plot("mag_img")
mask_plot, _ = self.make_image_plot("mask_img")
filtered_plot, _ = self.make_image_plot("filtered_img")
self.plot = HPlotContainer(meg_plot, mask_plot, filtered_plot)
# 创建套索工具
lasso_selection = LassoSelection(component=img)
lasso_overlay = LassoOverlay(lasso_selection = lasso_selection, component=img, selection_alpha=0.3)
img.tools.append(lasso_selection)
img.overlays.append(lasso_overlay)
self.lasso_selection = lasso_selection
# 监听套索工具的事件、开启时钟事件
lasso_selection.on_trait_change(self.lasso_updated, "disjoint_selections")
self.timer = Timer(50, self.on_timer)
img2 = cv.imread("fruits.jpg")
img_hsv2 = cv.Mat()
cv.cvtColor(img2, img_hsv2, cv.CV_BGR2HSV)
img_bp = cv.Mat()
cv.calcBackProject(cv.vector_Mat([img_hsv2]),
channels=channels,
hist=result,
backProject=img_bp,
ranges=ranges)
3 ###
img_th = cv.Mat()
cv.threshold(img_bp, img_th, 180, 255, cv.THRESH_BINARY)
4 ###
struct = np.ones((3, 3), np.uint8)
struct_mat = cv.asMat(struct, force_single_channel=True)
img_mp = cv.Mat()
cv.morphologyEx(img_th, img_mp, cv.MORPH_CLOSE, struct_mat, iterations=5)
import pylab as pl
import matplotlib.cm as cm
pl.subplot(231)
pl.imshow(img[:, :, ::-1])
pl.subplot(232)
pl.imshow(img2[:, :, ::-1])
pl.subplot(233)
pl.imshow(result[:], cmap=cm.gray)
pl.subplot(234)
pl.imshow(img_bp[:], cmap=cm.gray)
img2 = cv.imread("fruits.jpg")
img_hsv2 = cv.Mat()
cv.cvtColor(img2, img_hsv2, cv.CV_BGR2HSV)
img_bp = cv.Mat()
cv.calcBackProject(cv.vector_Mat([img_hsv2]),
channels=channels,
hist=result,
backProject=img_bp,
ranges = ranges)
3###
img_th = cv.Mat()
cv.threshold(img_bp, img_th, 180, 255, cv.THRESH_BINARY)
4###
struct = np.ones((3,3), np.uint8)
struct_mat = cv.asMat(struct, force_single_channel=True)
img_mp = cv.Mat()
cv.morphologyEx(img_th, img_mp, cv.MORPH_CLOSE, struct_mat, iterations=5)
import pylab as pl
import matplotlib.cm as cm
pl.subplot(231)
pl.imshow(img[:,:,::-1])
pl.subplot(232)
pl.imshow(img2[:,:,::-1])
pl.subplot(233)
pl.imshow(result[:], cmap=cm.gray)
pl.subplot(234)
pl.imshow(img_bp[:], cmap=cm.gray)
self.tmpbuf[:] = 0
for s in self.slice:
self.tmpbuf += self.w1[s]
self.tmpbuf /= 4
self.w2[1:-1, 1:-1] *= -1
self.w2[1:-1, 1:-1] += self.tmpbuf
self.w2 *= self.damping
self.w1, self.w2 = self.w2, self.w1
self.bmp[:, :, 0] = self.w1 + 128
self.bmp[:, :, 1] = self.bmp[:, :, 0] - (self.bmp[:, :, 0] >> 2)
self.bmp[:, :, 2] = self.bmp[:, :, 1]
return self.bmp
WIDTH, HEIGHT = 640, 480
video = cv.VideoWriter()
#video.open("waterwave.avi", cv.CV_FOURCC(*"DIB "), 30, cv.Size2i(WIDTH, HEIGHT))
video.open("waterwave.avi", cv.CV_FOURCC(*"ffds"), 30,
cv.Size2i(WIDTH, HEIGHT))
water = WaterWave(WIDTH, HEIGHT, 100, 0.97)
import time
start = time.clock()
for i in range(200):
if i % 30 == 0: print(i)
r = water.step()
mat = cv.asMat(r)
video << mat
del video
print((time.clock() - start))
# -*- coding: utf-8 -*-
import pyopencv as cv
import numpy as np
y, x = np.ogrid[-1:1:250j,-1:1:250j]
z = np.sin(10*np.sqrt(x*x+y*y))*0.5 + 0.5
np.round(z, decimals=1, out=z)
img = cv.asMat(z)
cv.namedWindow("demo1")
cv.imshow("demo1", img)
img2 = cv.Mat()
cv.Laplacian(img, img2, img.depth(), ksize=3)
cv.namedWindow("demo2")
cv.imshow("demo2", img2)
cv.waitKey(0)
# -*- coding: utf-8 -*-
import pyopencv as cv
import numpy as np
y, x = np.ogrid[-1:1:250j, -1:1:250j]
z = np.sin(10 * np.sqrt(x * x + y * y)) * 0.5 + 0.5
np.round(z, decimals=1, out=z)
img = cv.asMat(z)
cv.namedWindow("demo1")
cv.imshow("demo1", img)
img2 = cv.Mat()
cv.Laplacian(img, img2, img.depth(), ksize=3)
cv.namedWindow("demo2")
cv.imshow("demo2", img2)
cv.waitKey(0)
# 读入图片并缩小为1/2
img0 = cv.imread("lena.jpg")
size = img0.size()
w, h = size.width, size.height
img1 = cv.Mat()
cv.resize(img0, img1, cv.Size(w//2, h//2))
# 各种卷积核
kernels = [
(u"低通滤波器",np.array([[1,1,1],[1,2,1],[1,1,1]])*0.1),
(u"高通滤波器",np.array([[0,-1,0],[-1,5,-1],[0,-1,0]])),
(u"边缘检测",np.array([[-1,-1,-1],[-1,8,-1],[-1,-1,-1]]))
]
index = 0
for name, kernel in kernels:
plt.subplot(131+index)
# 将卷积核转换为Mat对象
kmat = cv.asMat(kernel.astype(np.float), force_single_channel=True)
img2 = cv.Mat()
cv.filter2D(img1, img2, -1, kmat)
# 由于matplotlib的颜色顺序和OpenCV的顺序相反
plt.imshow(img2[:,:,::-1])
plt.title(name)
index += 1
plt.gca().set_axis_off()
plt.subplots_adjust(0.02, 0, 0.98, 1, 0.02, 0)
plt.show()
def redraw(self):
img2 = cv.Mat()
kernel = cv.asMat(self.kernel*self.scale, force_single_channel=True)
cv.filter2D(self.img, img2, -1, kernel)
cv.imshow("Filter Demo", img2)
def affine(self):
self.img2 = cv.Mat()
M = cv.asMat(self.m, force_single_channel=True)
cv.warpAffine(self.img1, self.img2, M, self.img1.size(),
borderValue=cv.CV_RGB(255,255,255))
def Dilation(pos, user_data):
element = cv.asMat(np.ones((pos*2+1, pos*2+1), 'uint8'), True)
cv.dilate(src, dest, element)
cv.imshow("Erosion&Dilation window",dest);
def Dilation(pos, user_data):
element = cv.asMat(np.ones((pos * 2 + 1, pos * 2 + 1), 'uint8'), True)
cv.dilate(src, dest, element)
cv.imshow("Erosion&Dilation window", dest)
def redraw(self):
img2 = cv.Mat()
kernel = cv.asMat(self.kernel * self.scale, force_single_channel=True)
cv.filter2D(self.img, img2, -1, kernel)
cv.imshow("Filter Demo", img2)
# -*- coding: utf-8 -*-
import pyopencv as cv
import numpy as np
img = cv.imread("lena.jpg")
size = img.size()
w, h = size
img2 = cv.Mat()
map1, map2 = np.meshgrid(
np.linspace(0,w*2,w).astype(np.float32),
np.linspace(0,h*2,h).astype(np.float32),
)
map1 = cv.asMat(map1)
map2 = cv.asMat(map2)
cv.remap(img, img2, map1, map2, cv.INTER_LINEAR)
cv.namedWindow( "Remap Resize", cv.CV_WINDOW_AUTOSIZE )
cv.imshow("Remap Resize", img2)
cv.waitKey(0)
def new_dst(width=640, height=480, depth=3, mode='uint8'):
dst = cv.asMat(zeros((height, width, depth), mode))
return dst
def _get_morphology_element(v):
return cv.asMat(ones((v * 2 + 1, v * 2 + 1), 'uint8'), True)
# -*- coding: utf-8 -*-
import pyopencv as cv
import numpy as np
import time
img = cv.asMat(np.random.rand(1000, 1000))
row = cv.getGaussianKernel(7, -1)
col = cv.getGaussianKernel(5, -1)
kernel = cv.asMat(np.dot(col[:], row[:].T), force_single_channel=True)
img2 = cv.Mat()
img3 = cv.Mat()
start = time.clock()
cv.filter2D(img, img2, -1, kernel)
print(("filter2D:", time.clock() - start))
start = time.clock()
cv.sepFilter2D(img, img3, -1, row, col)
print(("sepFilter3D:", time.clock() - start))
print(("error=", np.max(np.abs(img2[:] - img3[:]))))
def asMat(src, *args, **kw):
return cv.asMat(src, *args, **kw)
def crop(src, x, y, w, h, mat=True):
# print y, y + h, x, x + w
v = src[y:y + h, x:x + w]
if mat:
v = cv.asMat(v)
return v
# -*- coding: utf-8 -*- import pyopencv as cv import numpy as np import time img = cv.asMat(np.random.rand(1000,1000)) row = cv.getGaussianKernel(7, -1) col = cv.getGaussianKernel(5, -1) kernel = cv.asMat(np.dot(col[:], row[:].T), force_single_channel=True) img2 = cv.Mat() img3 = cv.Mat() start = time.clock() cv.filter2D(img, img2, -1, kernel) print "filter2D:", time.clock() - start start = time.clock() cv.sepFilter2D(img, img3, -1, row, col) print "sepFilter3D:", time.clock() - start print "error=", np.max(np.abs(img2[:] - img3[:]))
def process(self, input_images, connected_outs):
if len(input_images) == 0:
return FAIL
src = input_images['Input']
dist_res = int(self.getParamContent('Distance resolution'))
angle_res = int(self.getParamContent('Angle resolution (degrees)'))
acc_thresh = int(self.getParamContent('Accumulator threshold'))
min_length = int(self.getParamContent('Minimum length'))
max_gap = int(self.getParamContent('Maximum gap'))
choice = self.getParamContent("Type of Hough transform")
if src.ndim > 2:
print "In '%s': The hough transform takes a binary image (or 8-bit) as input." % self.name
return FAIL
color_dst = numpy.empty((src.shape[0], src.shape[1], 3), dtype='uint8')
pycv.cvtColor(pycv.asMat(src), pycv.asMat(color_dst), pycv.CV_GRAY2BGR)
if choice == "Standard":
lines = pycv.HoughLines(pycv.asMat(src), dist_res,
pycv.CV_PI / angle_res, acc_thresh)
margin = 0.04
n = 8
pi = math.pi
h, w = src.shape[0:2]
for i in range(
min(len(lines),
int(self.getParamContent("draw # lines")))):
l = lines[i]
rho = l[0]
theta = l[1]
if theta > 3 * pi / 4: theta -= pi
if abs(rho) < w / n and abs(theta) < margin: pass
elif abs(rho) > w - w / n and abs(theta) < margin: pass
elif abs(rho) < h / n and abs(theta - pi / 2) < margin: pass
elif abs(rho) > h - h / n and abs(theta - pi / 2) < margin:
pass
else:
continue
a = math.cos(theta)
b = math.sin(theta)
x0 = a * rho
y0 = b * rho
pt1 = pycv.Point(int(round(x0 + 2000 * (-b))),
int(round(y0 + 2000 * (a))))
pt2 = pycv.Point(int(round(x0 - 2000 * (-b))),
int(round(y0 - 2000 * (a))))
pycv.line(
pycv.asMat(color_dst), pt1, pt2,
pycv.CV_RGB(random.randint(0, 255), random.randint(0, 255),
random.randint(0, 255)), 2, 8)
else:
lines = pycv.HoughLinesP(pycv.asMat(src), dist_res,
pycv.CV_PI / angle_res, acc_thresh,
min_length, max_gap)
for l in lines:
pycv.line(pycv.asMat(color_dst),
pycv.Point(int(l[0]), int(l[1])),
pycv.Point(int(l[2]), int(l[3])),
pycv.CV_RGB(*getRandColor()), 2, 8)
self.lines = [(item[0], item[1]) for item in lines]
return {
self.output_names[0]: color_dst,
self.output_names[1]: self.lines
}
def Closing(pos, user_data):
element = cv.asMat(np.ones((pos*2+1, pos*2+1), 'uint8'), True)
cv.dilate(src, image, element)
cv.erode(image, dest, element)
cv.imshow("Opening&Closing window",dest);
import pyopencv as cv
import numpy as np
import matplotlib.pyplot as plt
# 读入图片并缩小为1/2
img0 = cv.imread("lena.jpg")
size = img0.size()
w, h = size.width, size.height
img1 = cv.Mat()
cv.resize(img0, img1, cv.Size(w // 2, h // 2))
# 各种卷积核
kernels = [("低通滤波器", np.array([[1, 1, 1], [1, 2, 1], [1, 1, 1]]) * 0.1),
("高通滤波器", np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]])),
("边缘检测", np.array([[-1, -1, -1], [-1, 8, -1], [-1, -1, -1]]))]
index = 0
for name, kernel in kernels:
plt.subplot(131 + index)
# 将卷积核转换为Mat对象
kmat = cv.asMat(kernel.astype(np.float), force_single_channel=True)
img2 = cv.Mat()
cv.filter2D(img1, img2, -1, kmat)
# 由于matplotlib的颜色顺序和OpenCV的顺序相反
plt.imshow(img2[:, :, ::-1])
plt.title(name)
index += 1
plt.gca().set_axis_off()
plt.subplots_adjust(0.02, 0, 0.98, 1, 0.02, 0)
plt.show()
def Closing(pos, user_data):
element = cv.asMat(np.ones((pos * 2 + 1, pos * 2 + 1), 'uint8'), True)
cv.dilate(src, image, element)
cv.erode(image, dest, element)
cv.imshow("Opening&Closing window", dest)
