def inpaint(vid, threshold):
for i,img in enumerate(vid):
mask = img.copy()
mask[mask<threshold] = 0
cv2.inpaint(img, mask, 3, cv2.INPAINT_TELEA, img)
img[img>threshold] = 0
vid[i] = img
return vid
def inpaint_vessels(img,mask,radius,method):
img = cv2.imread(img)
mask = cv2.imread(mask,0)
kernel = np.ones((5,5),np.uint8)
mask = cv2.dilate(mask,kernel,iterations=1)
if method == 1:
dst = cv2.inpaint(img,mask,radius,cv2.INPAINT_TELEA)
else:
dst = cv2.inpaint(img,mask,radius,cv2.INPAINT_NS)
cv2.imshow('ori',img)
cv2.imshow('dst',dst)
cv2.imwrite('output_inpaint.jpg', dst)
cv2.waitKey(0)
cv2.destroyAllWindows()
def inpaint(self, image, topLeft=None, bottomRight=None,mask=None): if(mask is None): x1 = topLeft[0] x2 = bottomRight[0] y1 = topLeft[1] y2 = bottomRight[1] mask = numpy.zeros((image.shape[0], image.shape[1]), numpy.uint8) mask[y1:y2, x1:x2] = 1 image = cv2.inpaint(image, mask, 3, cv2.INPAINT_TELEA) else: image = cv2.inpaint(image, mask, 3, cv2.INPAINT_NS) return image
def main():
image = cv2.imread("../data/Damaged Image.tiff", 1)
mask_image = cv2.imread("../data/Mask.tiff", 0)
telea_image = cv2.inpaint(image, mask_image, 5, cv2.INPAINT_TELEA)
ns_image = cv2.inpaint(image, mask_image, 5, cv2.INPAINT_NS)
cv2.imshow("Orignal Image", image)
cv2.imshow("Mask Image", mask_image)
cv2.imshow("TELEA Restored Image", telea_image)
cv2.imshow("NS Restored Image", ns_image)
cv2.waitKey(0)
cv2.destroyAllWindows()
def run(self, meta):
mask = self.noise_mask(meta.data)
# Inpaint radius = 3, acceptable?
meta.data = cv2.inpaint(meta.data, numpy.array(mask, dtype=numpy.uint8), 5,
cv2.INPAINT_NS)
code.interact(local=vars())
return meta
def segment_image_inpaint(input_image, segdir, frpath):
path = os.path.join(segdir, frpath)
S = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
input_image = scipy.misc.imresize(input_image, np.shape(S))
input_image = cv2.inpaint(input_image, S, 5, cv2.INPAINT_NS)
input_image = input_image.astype(float) / np.max(input_image)
return input_image
def reflectionMask(img): # lab = cv2.cvtColor(img,cv2.COLOR_BGR2LAB) # l = lab[:,:,0] # a = lab[:,:,1] # b = lab[:,:,2] # ul = (l*2.56).astype(np.uint8) # ua = ((100+a)*1.28).astype(np.uint8) # ub = ((100+b)*1.28).astype(np.uint8) # ule = cv2.equalizeHist(ul) # ret, mask = cv2.threshold(ule, 242, 255, cv2.THRESH_BINARY) # mask = mask.astype(np.uint8) # iul = cv2.inpaint(ul, mask, 20, cv2.INPAINT_TELEA) # iua = cv2.inpaint(ua, mask, 20, cv2.INPAINT_TELEA) # iub = cv2.inpaint(ub, mask, 20, cv2.INPAINT_TELEA) # ifl = (iul.astype(np.float32))/2.56 # ifa = (iua.astype(np.float32)-128)/1.28 # ifb = (iub.astype(np.float32)-128)/1.28 # lab = np.array([ifl,ifa,ifb]).swapaxes(0,1).swapaxes(1,2) b = img[:,:,0] g = img[:,:,1] r = img[:,:,2] ret, mask = cv2.threshold(2*b+g+r,2.8,255, cv2.THRESH_BINARY) mask = mask.astype(np.uint8) mimg = cv2.inpaint((img*256).astype(np.uint8),mask, 10, cv2.INPAINT_TELEA) mimg = mimg.astype(np.float32)/256 return (mimg,mask)
def main():
import sys
try:
fn = sys.argv[1]
except:
fn = 'fruits.jpg'
img = cv.imread(cv.samples.findFile(fn))
if img is None:
print('Failed to load image file:', fn)
sys.exit(1)
img_mark = img.copy()
mark = np.zeros(img.shape[:2], np.uint8)
sketch = Sketcher('img', [img_mark, mark], lambda : ((255, 255, 255), 255))
while True:
ch = cv.waitKey()
if ch == 27:
break
if ch == ord(' '):
res = cv.inpaint(img_mark, mark, 3, cv.INPAINT_TELEA)
cv.imshow('inpaint', res)
if ch == ord('r'):
img_mark[:] = img
mark[:] = 0
sketch.show()
print('Done')
def preprocess_image(image):
# Copy the depth part of the image
depth_pixels = image.pixels[..., 2].copy()
depth_pixels = rescale_to_opencv_image(depth_pixels)
filtered_depth_pixels = median_filter(depth_pixels, 5)
# Build mask for floodfilling, this lets me ignore all the pixels
# from the background and around the ears
mask = np.zeros((depth_pixels.shape[0] + 2, depth_pixels.shape[1] + 2),
dtype=np.uint8)
# Flood fill from top left
cv2.floodFill(filtered_depth_pixels, mask, (0, 0),
(255, 255, 255), flags=cv2.FLOODFILL_MASK_ONLY)
# Flood fill from top right
cv2.floodFill(filtered_depth_pixels, mask, (depth_pixels.shape[1] - 1, 0),
(255, 255, 255), flags=cv2.FLOODFILL_MASK_ONLY)
# Truncate and negate the flood filled areas to find the facial region
floodfill_mask = (~mask.astype(np.bool))[1:-1, 1:-1]
# Build a mask of the areas inside the face that need inpainting
inpaint_mask = ~image.mask.mask & floodfill_mask
# Inpaint the image and filter to smooth
inpainted_pixels = cv2.inpaint(depth_pixels,
inpaint_mask.astype(np.uint8),
5, cv2.INPAINT_NS)
inpainted_pixels = median_filter(inpainted_pixels, 5)
# Back to depth pixels
image.pixels[..., 2] = rescale_to_depth_image(image, inpainted_pixels)
# Reset the mask!
image.mask.pixels[..., 0] = ~np.isnan(image.pixels[..., 2])
def Inpaintv1(Depth):
Depth_Small = Depth
Temp2 = Depth
x1 = int(Depth.shape[0] * 0.2)
x2 = int(Depth.shape[1] * 0.2)
x3 = Depth.shape[2]
cv2.resize(Depth, (x1, x2), Depth_Small)
Temp = Depth_Small
mask = (Depth_Small == 0)
zeros = np.zeros(Depth_Small.shape, Depth_Small.dtype)
ones = np.ones(Depth_Small.shape, Depth_Small.dtype)
ones *= 255
maskk = np.where(mask == True, ones, zeros)
maskk = maskk[:, :, 0]
cv2.inpaint(Depth_Small, maskk, 10.0, cv2.INPAINT_TELEA, Temp)
cv2.resize(Temp, (Depth.shape[0], Depth.shape[1]), Temp2)
return Temp2
def depth_features(depth_img):
'''
fill in np.nan holes using cv2.inpaint,
then return histogram of oriented gradients
'''
mask = np.isnan(depth_img).astype(np.uint8)
normed = (255.0 * depth_img / np.nanmax(depth_img)).astype(np.uint8)
filled = cv2.inpaint(normed, mask, 8, cv2.INPAINT_NS)
hogged = hog(filled, pixels_per_cell=(40,40), cells_per_block=(2,2))
return hogged
def inpaint(p_image, p_mask):
# if len(sys.argv) < 3:
# print "Not enough params"
# sys.exit()
img = cv2.imread(p_image)
mask = cv2.imread(p_mask, 0)
res = cv2.inpaint(img, mask, 3, cv2.INPAINT_NS)
cv2.imshow('res', res)
cv2.waitKey(10000)
cv2.destroyAllWindows()
sys.exit()
def remove_glare(im):
# generate binary image mask - dilated circles around the saturated bright spots at the center
temp = im[y-w:y+w, x-w:x+w,1] # single channel
ret, temp_mask = cv2.threshold(temp, thresh*256, 255, cv2.THRESH_BINARY)
kernel = numpy.ones((25,25), 'uint8')
temp_mask = cv2.dilate(temp_mask, kernel)
# perform the inpainting...
im[y-w:y+w, x-w:x+w,:] = cv2.inpaint(im[y-w:y+w, x-w:x+w,:], temp_mask, 1, cv2.INPAINT_TELEA)
# return file
return im
def replace(self, marker): firstPoint = self.refinePoint(marker.getFirstPoint()) secondPoint = self.refinePoint(marker.getSecondPoint()) x1 = firstPoint[0] y1 = firstPoint[1] x2 = secondPoint[0] y2 = secondPoint[1] mask = numpy.zeros((self.image.shape[0], self.image.shape[1]), numpy.uint8) mask[y1:y2, x1:x2] = 1 self.image = cv2.inpaint(self.image, mask, 3, cv2.INPAINT_TELEA)
def erase_specular(image, lower_threshold=0.0, upper_threshold=150.0):
"""erase_specular: removes specular reflections
within given threshold using a binary mask (hi_mask)
"""
thresh = cv2.inRange(image, np.asarray(float(lower_threshold)), np.asarray(256.0))
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7, 7))
hi_mask = cv2.dilate(thresh, kernel, iterations=2)
specular = cv2.inpaint(image, hi_mask, 2, flags=cv2.INPAINT_TELEA)
# return cv2.max(hi_mask,image)
return specular
def _transform_msg_into_cv2images(self, msg_color, msg_mapping, msg_mask, inpaint):
rgb_frame_numpy = numpy.fromstring(msg_color, numpy.uint8).reshape(1080, 1920,2)
frame_rgb = cv2.cvtColor(rgb_frame_numpy, cv2.COLOR_YUV2BGR_YUY2) # YUY2 to BGR
mapped_frame_numpy = numpy.fromstring(msg_mapping, numpy.uint8).reshape(1080*self._factor, 1920*self._factor)
mapped_image = numpy.uint8(cv2.normalize(mapped_frame_numpy, None, 0, 255, cv2.NORM_MINMAX))
mask_numpy = numpy.fromstring(msg_mask, numpy.uint8).reshape(1080*self._factor, 1920*self._factor)
mask = numpy.uint8(cv2.normalize(mask_numpy, None, 0, 255, cv2.NORM_MINMAX))
if inpaint is True:
smoothing = cv2.inpaint(mapped_image, mask, 3, cv2.INPAINT_TELEA)
return frame_rgb, mapped_image, mask
def draw(self):
if self.mask_artist is None:
self.draw_image(self.img)
return
mask = self.mask_artist.get_mask_array()
if self.img.shape[:2] == mask.shape:
img2 = cv2.inpaint(self.img, mask, self.r, getattr(cv2, self.method))
self.img2 = img2
self.show_mask = False
self.mask_artist.hide_mask()
self.draw_image(img2)
else:
self.draw_image(self.img)
def inpaint_image(img_file):
img = cv2.imread(img_file)
neg_mask = np.zeros(img.shape, dtype=np.uint8)
text_area = np.array([ [110,450], [440,450], [440,690], [110,690] ], np.int32)
cv2.fillConvexPoly(neg_mask, text_area, (255,255,255))
neg_mask = np.invert(neg_mask)
mask = np.bitwise_or(img, neg_mask)
gray = cv2.cvtColor(mask, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray,0,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
blank = cv2.inpaint(img,thresh,3,cv2.INPAINT_TELEA)
blank_file = 'photos/blank_'+img_file.split('/')[1]
cv2.imwrite(blank_file, blank)
return blank_file
def image_callback(self, image_in):
""" Get image to which we're subscribed. """
# Import and convert
image_cv2 = self.rcv.toCv2(image_in)
image_hsv = cv2.cvtColor(image_cv2, cv2.COLOR_BGR2HSV)
try:
pink_lowerb = numpy.array((140, 100,100))
pink_upperb = numpy.array((170,255, 255))
pink_x, pink_y, pink_area = self.rcv.find_marker(image_cv2, pink_lowerb, pink_upperb)
green_lowerb = numpy.array((50, 100,100))
green_upperb = numpy.array((80,255, 255))
green_x, green_y, green_area = self.rcv.find_marker(image_cv2, green_lowerb, green_upperb)
special_area = image_hsv[pink_y:green_y, pink_x:green_x]
markings = image_cv2[pink_y:green_y, pink_x:green_x]
markings = cv2.cvtColor(markings, cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(markings, 10, 20)
img_height = len(image_cv2)
img_width = len(image_cv2[0])
mask = numpy.zeros((img_height, img_width), dtype=numpy.uint8)
mask[pink_y:green_y, pink_x:green_x] = edges
kernel = numpy.ones((5,5),'uint8')
mask = cv2.dilate(mask, kernel)
# mask = cv2.erode(mask, kernel)
board_depth = self.depth_image[pink_y, pink_x]
# print "board depth = {0}".format(board_depth)
# print self.depth_image
# print numpy.where(self.depth_image <= board_depth - 0.2)
# http://stackoverflow.com/questions/432112/is-there-a-numpy-function-to-return-the-first-index-of-something-in-an-array
# for i in range(img_height):
# for j in range(img_width):
# if self.depth_image[i][j] <= board_depth - 0.25:
# mask[i][j] = 0
image_cv2 = cv2.inpaint(image_cv2, mask, 5, cv2.INPAINT_TELEA)
# cv2.rectangle(image_cv2, (green_x, green_y), (pink_x, pink_y), (0, 0, 0), 3)
except(ZeroDivisionError):
pass
# except(ZeroDivisionError, TypeError, AttributeError):
self.rcv.imshow(self.depth_image)
# self.rcv.imshow(image_cv2)
# Convert back to ROS Image msg
image_out = self.rcv.toRos(image_cv2)
self.pub.publish(image_out)
def inpaint_retina(source_path,mask_path,dest_path,img_ext):
print "Hello"
sources = [ f for f in listdir(source_path) if f.endswith(img_ext) ]
print "Number of files to process is ",len(sources),"."
masks = [ f for f in listdir(mask_path) if f.endswith(img_ext) ]
kernel = np.ones((7,7),np.uint8)
for (s,m) in zip(sources,masks):
img = cv2.imread(join(source_path,s))
mask = cv2.imread(join(mask_path,m),0)
mask = cv2.threshold(mask,0, 255, cv2.THRESH_BINARY)[1]
mask = cv2.dilate(mask,kernel,iterations = 1)
dst = cv2.inpaint(img,mask,55,cv2.INPAINT_NS)
cv2.imwrite(join(dest_path,s),dst)
print "File ",s, " processed.\n"
return
def LightInpaint(im, mask, rsizFac = 4): maskGlints = cv2.resize(mask, (im.shape[1]/rsizFac, im.shape[0]/rsizFac)) imTmp = cv2.resize(im, (im.shape[1]/rsizFac, im.shape[0]/rsizFac)) #imTmp = im.copy() krnl = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5,5)) dilt = cv2.dilate(maskGlints, krnl) inpaintGlint = np.float64(cv2.inpaint(np.uint8(imTmp*255),np.uint8(dilt*255),10,cv2.INPAINT_NS))/255 fullInpaint = cv2.resize(inpaintGlint, (im.shape[1], im.shape[0]), interpolation = cv2.INTER_CUBIC) rsizDilt = cv2.resize(dilt, (im.shape[1], im.shape[0]), interpolation = cv2.INTER_NEAREST) rsizDilt[rsizDilt!=0]=1 fullInpaintMask = fullInpaint*rsizDilt tmpMask = np.zeros_like(im) tmpMask[np.where(rsizDilt==0)]=1 imNew = im*tmpMask + fullInpaintMask return imNew
def inpaint(img, threshold=50, minLineLength=50, maxLineGap=5, thickness=2, inpaintRadius=3):
edges = cv2.Canny(img, 50, 150, apertureSize=3)
# 经验参数
lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold, minLineLength=minLineLength, maxLineGap=maxLineGap)
if lines is None:
return img
result = np.zeros(img.shape[:2], np.uint8)
count = 0
h, w = img.shape[:2]
for x1, y1, x2, y2 in lines[0]:
if y1 + 2 <= h:
y1 = y1 + 2
if y2 + 2 <= h:
y2 = y2 + 2
cv2.line(img, (x1, y1), (x2, y2), (255, 255, 255), thickness)
return cv2.inpaint(img, result, inpaintRadius, cv2.INPAINT_TELEA)
def erase_specular(self, eye_img, debug=False):
# Rather arbitrary decision on how large a specularity may be
max_specular_contour_area = sum(eye_img.shape[:2])/2
# Extract top 50% of intensities
eye_img_grey = cv2.cvtColor(eye_img, cv2.COLOR_BGR2GRAY)
eye_img_grey_blur = cv2.GaussianBlur(eye_img_grey, (5, 5), 0)
# Close to suppress eyelashes
morph_kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5))
eye_img_grey_blur = cv2.morphologyEx(eye_img_grey_blur, cv2.MORPH_CLOSE, morph_kernel)
if eye_img_grey_blur is None:
raise eye_extractor.NoEyesFound()
thresh_val = int(np.percentile(eye_img_grey_blur, 50))
_, thresh_img = cv2.threshold(eye_img_grey_blur, thresh_val, 255, cv2.THRESH_BINARY)
# Find all contours and throw away the big ones
contours, _ = cv2.findContours(np.copy(thresh_img), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
small_contours = filter(lambda x : cv2.contourArea(x) < max_specular_contour_area, contours)
small_contours_mask = np.zeros_like(eye_img_grey)
cv2.drawContours(small_contours_mask, small_contours, -1, 255, -1)
# Dilate the smallest contours found
small_contours_mask_dilated = cv2.dilate(small_contours_mask, morph_kernel)
removed_specular_img = cv2.inpaint(eye_img, small_contours_mask_dilated, 2, flags=cv2.INPAINT_TELEA)
if debug:
thresh_hierarchy = cv2.cvtColor(eye_img_grey, cv2.COLOR_GRAY2BGR)
cv2.drawContours(thresh_hierarchy, contours, -1, (0, 0, 255), -1)
thresh_hierarchy = cv2.add(thresh_hierarchy, cv2.cvtColor(small_contours_mask_dilated, cv2.COLOR_GRAY2BGR))
cv2.drawContours(thresh_hierarchy, small_contours, -1, (255, 0, 0), -1)
stacked_imgs = np.concatenate([eye_img, thresh_hierarchy, removed_specular_img],axis=1)
if debug == 1:
self.full_debug_img = stacked_imgs
elif debug == 2:
self.full_debug_img = image_utils.stack_imgs_vertical([self.full_debug_img, stacked_imgs])
cv2.imshow(winname, self.full_debug_img)
elif debug == 3:
cv2.imshow(winname, stacked_imgs);
return removed_specular_img
def _calc_block_mean_variance(image, mask, blocksize):
"""Adaptively determines image background.
Args:
image: image converted 1-channel image.
mask: 1-channel mask, same size as image.
blocksize: adaptive algorithm parameter.
Returns:
image of same size as input with foreground inpainted with background.
"""
I = image.copy()
I_f = I.astype(np.float32) / 255. # Used for mean and std.
result = np.zeros(
(image.shape[0] / blocksize, image.shape[1] / blocksize),
dtype=np.float32)
for i in xrange(0, image.shape[0] - blocksize, blocksize):
for j in xrange(0, image.shape[1] - blocksize, blocksize):
patch = I_f[i:i+blocksize+1, j:j+blocksize+1]
mask_patch = mask[i:i+blocksize+1, j:j+blocksize+1]
tmp1 = np.zeros((blocksize, blocksize))
tmp2 = np.zeros((blocksize, blocksize))
mean, std_dev = cv2.meanStdDev(patch, tmp1, tmp2, mask_patch)
value = 0
if std_dev[0][0] > MEAN_VARIANCE_THRESHOLD:
value = mean[0][0]
result[i/blocksize, j/blocksize] = value
small_image = cv2.resize(I, (image.shape[1] / blocksize,
image.shape[0] / blocksize))
res, inpaintmask = cv2.threshold(result, 0.02, 1, cv2.THRESH_BINARY)
inpainted = cv2.inpaint(small_image, inpaintmask.astype(np.uint8), 5,
cv2.INPAINT_TELEA)
res = cv2.resize(inpainted, (image.shape[1], image.shape[0]))
return res
def inpaint(image, mask, flags=cv2.INPAINT_NS):
### Added to Wralea.py
"""The function can be used to restore the image content in a provided
region of interest using the neighborhood to this region.
:Parameters:
-`image` specifies the loaded image that should be processsed.
The image can be either a 8-Bit 1- or 3-channel image.
-`mask` specifies a mask by a provided 1-channel 8-Bit image, in which
non-zero pixels define the regions of interest, which should be restored.
-`flags`
:Returns:
-`restoredimage`, which is the image with the restored pixels.
:Notes:
"""
return cv2.inpaint(image, mask, flags)
def pretty_depth(frame): '''Assumes a very raw 16 bit depth image from the kinect. Converts to an 8 bit image, keeping as much contrast as possible.''' frame = frame.copy() errs = np.where(frame==2047) # constrain the frame to 0-255 frame = np.clip(frame, 768, 1023) frame = frame - 768 frame = np.uint8(frame) # ok, now we have a uint8 object to deal with # now let's fix all the errors using inpainting mask = np.zeros(frame.shape, dtype=np.uint8) mask[errs] = 1 inpainted = cv2.inpaint(frame,mask,2,cv2.INPAINT_TELEA) # there are also some errors around the edge of the image which dont get dealt with cropped = inpainted[1:-1, 1:-1] padded = cv2.copyMakeBorder(cropped,1,1,1,1,cv2.BORDER_REPLICATE) return padded
def clean_depth_image(self, depth_image):
"""
http://www.morethantechnical.com/2011/03/05/neat-opencv-smoothing-trick-when-kineacking-kinect-hacking-w-code/
need to get a mask for inpaintMask
"""
depth_image = depth_image.copy()
depth_image[depth_image > MAX_DEPTH] = MAX_DEPTH + BUFFER_DEPTH
depth_image = (depth_image / (MAX_DEPTH + BUFFER_DEPTH)) * 255
depth_image = depth_image.astype(np.uint8)
mask = np.zeros_like(depth_image)
mask[depth_image == 0] = 255
mask = cv2.resize(mask, (320, 240))
depth_image = cv2.resize(depth_image, (320, 240))
depth_image = cv2.inpaint(depth_image, mask, 5.0, cv2.INPAINT_TELEA)
return depth_image
data = json.load(f)
db = soccer3d.YoutubeVideo(opt.path_to_data)
db.digest_metadata()
img = db.get_frame(0)
mask = db.get_mask_from_detectron(0)
cam_npy = db.calib[db.frame_basenames[0]]
cam = cam_utils.Camera('tmp', cam_npy['A'], cam_npy['R'], cam_npy['T'], db.shape[0], db.shape[1])
mask = cv2.dilate(mask, np.ones((9, 9), np.int8), iterations=1)
img = cv2.inpaint((img[:, :, (2, 1, 0)]*255).astype(np.uint8), (mask*255).astype(np.uint8), 3, cv2.INPAINT_TELEA)[:, :, (2, 1, 0)]/255.
W, H = 104.73, 67.74
# a X-------------------------X b
# | |
# | |
# d X-------------------------X c
# Whole field
p3 = np.array([[-W/2., 0, H/2], [W/2., 0, H/2], [W/2., 0, -H/2] , [-W/2., 0, -H/2]])
p2, _ = cam.project(p3)
pp2 = np.array([[0, 0], [db.shape[1]-1, 0], [db.shape[1], db.shape[0]], [0, db.shape[0]-1]])
filled = (img*255).astype(np.uint8)
def LaplacianDeform(depth_in, handle_3d, intrinsic=None, vis=False):
"""depth_ctrl is a reference Nx3 depth map
Args:
depth_in: the depth predicted by network
handle_3d: the control sparse points, in [x, y, z]
intrinsic: the intrinsic matrix
vis: whether visualize using igl
Output:
depth: the warpped depth through asap
"""
height, width = depth_in.shape[0], depth_in.shape[1]
depth_in, y, x, handle_3d = ReScale(depth_in,
handle_3d,
intrinsic,
get_image_coor=True)
point_3d = uts_3d.depth2xyz(depth_in, intrinsic, False)
select_id = y * width + x
# test_id = range(10)
# select_id = select_id[test_id]
one_hot = np.zeros((point_3d.shape[0]), dtype=np.int32)
one_hot[select_id] = 1
mesh_idx = uts_3d.grid_mesh(height, width)
V = igl.eigen.MatrixXd(np.float64(point_3d))
U = V
F = igl.eigen.MatrixXi(np.int32(mesh_idx))
S = igl.eigen.MatrixXd(np.float64(one_hot))
b = igl.eigen.MatrixXi(np.int32(select_id))
P_origin = igl.eigen.MatrixXd(np.float64(point_3d[select_id, :]))
P = igl.eigen.MatrixXd(np.float64(handle_3d))
bc = igl.eigen.MatrixXd(np.float64(handle_3d))
arap_data = igl.ARAPData()
# Set color based on selection
# C = igl.eigen.MatrixXd(F.rows(), 3)
# purple = igl.eigen.MatrixXd([[80.0 / 255.0, 64.0 / 255.0, 255.0 / 255.0]])
# gold = igl.eigen.MatrixXd([[255.0 / 255.0, 228.0 / 255.0, 58.0 / 255.0]])
# pdb.set_trace()
# for f in range(0, F.rows()):
# if S[F[f, 0]] > 0 or S[F[f, 1]] > 0 or S[F[f, 2]] > 0:
# C.setRow(f, purple)
# else:
# C.setRow(f, gold)
# # Plot the mesh with pseudocolors
# viewer = igl.viewer.Viewer()
# viewer.data.set_mesh(V, F)
# viewer.data.set_colors(C)
# viewer.core.is_animating = False
# viewer.launch()
if vis:
viewer = igl.viewer.Viewer()
viewer.data.set_mesh(U, F)
viewer.data.add_points(P, igl.eigen.MatrixXd([[1, 0, 0]]))
viewer.core.is_animating = False
viewer.launch()
# start compute deform
arap_data.max_iter = 30
arap_data.ym = 450
igl.arap_precomputation(V, F, V.cols(), b, arap_data)
igl.arap_solve(bc, arap_data, U)
if vis:
viewer = igl.viewer.Viewer()
viewer.data.set_mesh(V, F)
# viewer.data.add_points(P_origin, igl.eigen.MatrixXd([[0, 0, 1]]))
# viewer.data.add_points(P, igl.eigen.MatrixXd([[0, 1, 0]]))
viewer.core.is_animating = False
viewer.launch()
point_3d_new = np.float32(np.array(U, dtype='float64', order='C'))
depth = uts_3d.xyz2depth(point_3d_new, intrinsic, depth_in.shape)
mask = depth <= 0
max_depth = np.max(depth)
depth_inpaint = cv2.inpaint(np.uint8(depth / max_depth * 255),
np.uint8(mask), 5, cv2.INPAINT_TELEA)
depth[mask] = np.float32(depth_inpaint[mask]) * max_depth / 255
return depth
# ret,thresh = cv2.threshold(img_gray,100,255,cv2.THRESH_BINARY_INV)
# ostu not work
# ret3, thresh_ = cv2.threshold(img_gray, 0, 255, cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
# contours, hierarchy = cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
# cv2.drawContours(thresh,contours,-1,(255,255,255),5)
mask = cv2.cvtColor(thresh, cv2.COLOR_GRAY2RGB)
kernel = np.ones((5,5),np.uint8)
dilation = cv2.dilate(mask,kernel,iterations = 2)
cv2.imwrite('mask.jpg', dilation)
########################## Inpaint Function ##########################################
mask = cv2.imread('mask.jpg', 0)
dst_NS = cv2.inpaint(img, mask, 5, cv2.INPAINT_NS)
cv2.imwrite('inpaint.jpg', dst_NS)
# cv2.imwrite('in.jpg', dst_TELEA)
# plt.subplot(221),plt.imshow(img_gray)
# # plt.title('Original Image')
# plt.subplot(222),plt.imshow(thresh,'gray')
# # plt.title('Extracted extra region Image')
# plt.subplot(223),plt.imshow(dst_NS)
# # plt.title('Cropped Image')
# plt.show()
#########################
# Prepare the stuff
#########################
# read images and create mask
rgbImg = cv2.imread(rgbImgPath)
depImg = cv2.imread(depImgPath, cv2.IMREAD_UNCHANGED)
# inpainting
scaleOri = np.amax(depImg)
inPaiMa = np.where(depImg == 0.0, 255, 0)
inPaiMa = inPaiMa.astype(np.uint8)
inPaiDia = 5.0
depth_refine = depImg.astype(np.float32)
depPaint = cv2.inpaint(depth_refine, inPaiMa, inPaiDia, cv2.INPAINT_NS)
depNorm = depPaint - np.amin(depPaint)
rangeD = np.amax(depNorm)
depNorm = np.divide(depNorm, rangeD)
depth_refine = np.multiply(depNorm, scaleOri)
# create image number and name
template = '00000'
s = int(s)
ssm = int(ss) + 1
pre = (s-1) * 1296
img_id = pre + ssm
tempSS = template[:-len(str(img_id))]
#coding=utf-8
#图片修复
import cv2
import numpy as np
path = "img/inpaint.png"
img = cv2.imread(path)
hight, width, depth = img.shape[0:3]
#图片二值化处理，把[240, 240, 240]~[255, 255, 255]以外的颜色变成0
thresh = cv2.inRange(img, np.array([240, 240, 240]), np.array([255, 255, 255]))
#创建形状和尺寸的结构元素
kernel = np.ones((3, 3), np.uint8)
#扩张待修复区域
hi_mask = cv2.dilate(thresh, kernel, iterations=1)
specular = cv2.inpaint(img, hi_mask, 5, flags=cv2.INPAINT_TELEA)
cv2.namedWindow("Image", 0)
cv2.resizeWindow("Image", int(width / 2), int(hight / 2))
cv2.imshow("Image", img)
cv2.namedWindow("newImage", 0)
cv2.resizeWindow("newImage", int(width / 2), int(hight / 2))
cv2.imshow("newImage", specular)
cv2.waitKey(0)
cv2.destroyAllWindows()
mask = cv.resize(mask, (513, 288), interpolation=cv.INTER_AREA)
video = cv.VideoWriter('lion_defenced.mp4', cv.VideoWriter_fourcc(*'MP4V'), 30,
(513, 288))
while (cap.isOpened()):
ret, frame = cap.read()
if ret == True:
frame = cv.resize(frame, (513, 288), interpolation=cv.INTER_AREA)
# dst = cv.inpaint(frame,mask,3,cv.INPAINT_TELEA)
dst = cv.inpaint(frame, mask, 3, cv.INPAINT_NS)
video.write(dst)
key = cv.waitKey(30)
if key == 27 or key == 'q':
cont = False
break
else:
break
imgThresholdInv = cv2.bitwise_not(imgGreenThreshold)
imgThresholdColor = cv2.cvtColor(imgGreenThreshold, cv2.COLOR_GRAY2BGR)
imgDiff = cv2.add(imgInput, imgThresholdColor)
imgDiffHSV = cv2.cvtColor(imgDiff, cv2.COLOR_BGR2HSV)
imgGray = imgDiffHSV[:, :, 2]
cv2.imshow("imgGray", imgGray)
cv2.imshow("imgGreenThreshold", imgGreenThreshold)
r_img = m_img = np.array(imgGray)
rimg = spc.derive_m(imgInput, r_img)
s_img = spc.derive_saturation(imgInput, rimg)
spec_mask = spc.check_pixel_specularity(rimg, s_img)
cv2.imshow("spec_mask", spec_mask)
enlarged_spec = spc.enlarge_specularity(spec_mask)
radius = 12
telea = cv2.inpaint(imgInput, enlarged_spec, radius, cv2.INPAINT_TELEA)
ns = cv2.inpaint(imgInput, enlarged_spec, radius, cv2.INPAINT_NS)
cv2.imshow("telea", telea)
cv2.imshow("ns", ns)
#imgBlurred = cv2.GaussianBlur(imgGray, (3, 3), 0)
#ret, imgThresh = cv2.threshold(imgBlurred, 250, 255, cv2.THRESH_BINARY_INV)
#kernel = np.ones((3,3), np.uint8)
#dist = cv2.distanceTransform(imgGreenThreshold, cv2.DIST_L2, cv2.DIST_MASK_PRECISE)
#dist = cv2.normalize(dist, None, 255,0, cv2.NORM_MINMAX, cv2.CV_8UC1)
#cv2.imshow("dist", dist)
#kernel = np.ones((3,3), np.uint8)
#morph = cv2.morphologyEx(imgThresh, cv2.MORPH_CLOSE, kernel)
#morph = cv2.morphologyEx(morph, cv2.MORPH_OPEN, kernel)
#Resize
res = cv2.resize(img, None, fx=0.2, fy=0.2, interpolation=cv2.INTER_CUBIC)
#Erode
kernel = np.ones((5, 5), np.uint8)
imgEroded = cv2.erode(res, kernel, iterations=1)
#Convert to grayscale
gray = cv2.cvtColor(imgEroded, cv2.COLOR_BGR2GRAY)
#Binarize to get highlight
ret, mask = cv2.threshold(gray, 230, 255, cv2.THRESH_BINARY)
#Inpaint highlight
dst = cv2.inpaint(imgEroded, mask, 3, cv2.INPAINT_TELEA)
#Convert to HSV
HSV = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
hsv = cv2.cvtColor(dst, cv2.COLOR_BGR2HSV)
while True:
# img = cv2.imread("lambo.png")
h_min = cv2.getTrackbarPos("Hue Min", "TrackBars")
h_max = cv2.getTrackbarPos("Hue Max", "TrackBars")
s_min = cv2.getTrackbarPos("Sat Min", "TrackBars")
s_max = cv2.getTrackbarPos("Sat Max", "TrackBars")
v_min = cv2.getTrackbarPos("Val Min", "TrackBars")
v_max = cv2.getTrackbarPos("Val Max", "TrackBars")
print(h_min, h_max, s_min, s_max, v_min, v_max)
def inpaint(img, mask):
res = cv.inpaint(img, mask, 129, cv.INPAINT_NS)
cv.imshow('result', res)
cv.waitKey(800)
return res
import numpy as np
import cv2
img = cv2.imread('new2.jpg')
img_rgb=cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_hsv=cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
img_hls=cv2.cvtColor(img, cv2.COLOR_BGR2HLS)
hsv_gray = cv2.cvtColor(img_hsv,cv2.COLOR_BGR2GRAY)
kernel_size = 5
hsv_blur = cv2.GaussianBlur(hsv_gray,(kernel_size, kernel_size),0)
ret, thresh = cv2.threshold(hsv_gray, 150, 255, cv2.THRESH_BINARY)
kernel = np.ones((2,2))
erosion = cv2.dilate(thresh,kernel,iterations = 1)
final = cv2.inpaint(img, thresh, 3, cv2.INPAINT_TELEA)
cv2.imshow('hsv_gray', hsv_gray)
cv2.imshow('img', img)
cv2.imshow('final', final)
cv2.imshow('erosion', erosion)
cv2.imshow('img_hsv', img_hsv)
#cv2.imshow('img_hls', img_hls)
cv2.waitKey(0)
cv2.destroyAllWindows()
def inpaint(image, mask):
k = numpy.ones((5, 5), numpy.uint8)
m = cv2.dilate(mask, k, iterations=1)
i = cv2.inpaint(image, m, 10, cv2.INPAINT_TELEA)
return i
import numpy as np
import cv2
img = cv2.imread("images/Messi.jpg")
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY);
scratch = cv2.imread("images/Messi_scratches.jpg")
scratch_gray = cv2.cvtColor(scratch, cv2.COLOR_BGR2GRAY);
cv2.imshow("Scratch", scratch)
mask = scratch - img
abs_diff = cv2.absdiff(scratch_gray, img_gray)
# cv2.imshow("Abs Diff", abs_diff)
_, mask = cv2.threshold(abs_diff, 100, 255, cv2.THRESH_BINARY)
# cv2.imshow("Mask", mask)
dst = cv2.inpaint(img, mask, 3, cv2.INPAINT_TELEA)
cv2.imshow('dst', dst)
cv2.waitKey(0)
cv2.destroyAllWindows()
img_blur = cv2.GaussianBlur(img_RGB_after, (5, 5), 0)
mask = np.zeros(img_RGB_after.shape, np.uint8)
gray = cv2.cvtColor(img_RGB_after, cv2.COLOR_BGR2GRAY)
# laplacian = cv2.Laplacian(gray,cv2.CV_64F)
#
# cv2.imshow("laplacian", laplacian)
thresh = cv2.threshold(gray, 60, 255, cv2.THRESH_BINARY_INV)[1]
# thresh = cv2.threshold(gray, 60, 255, cv2.THRESH_OTSU)[1]
# cv2.imshow("tresholded image", thresh)
inpaint = cv2.inpaint(img_RGB_after, thresh, 3, cv2.INPAINT_TELEA)
# cv2.imshow("inpaint image", inpaint)
kernel = np.ones((6, 6), np.uint8)
closing = cv2.bitwise_not(thresh)
closing = cv2.morphologyEx(closing, cv2.MORPH_CLOSE, kernel)
closing = cv2.dilate(closing, None, iterations=2)
closing = cv2.erode(closing, None, iterations=2)
clop = np.where(closing[..., None] == 0, [255, 255, 255], [0, 0, 0])
cv2.imshow("clop img", closing)
if __name__ == '__main__':
#filename = os.path.join('prova.jpg')
#image_orig = io.imread(filename)
xmin = 351 #xmin
ymin = 52 #ymin
xmax = 598
ymax = 140
img = cv.imread('prova.jpg')
h, w = img.shape[:-1]
# Create mask with a bounding box defect regions
#mask= np.zeros(img.shape[:-1])
cvmask = np.zeros((h, w, 1), dtype=np.uint8)
cvmask[ymin:ymin + ymax, xmin:xmin + xmax] = 1
#image_result = inpaint.inpaint_biharmonic(image_defect, mask, multichannel=True)
image_result = cv.inpaint(img, cvmask, 3, cv.INPAINT_TELEA)
#cv.imshow('dst', image_result)
image_result2 = cv.inpaint(img, cvmask, 3, cv.INPAINT_NS)
#cv.imshow('dst', image_result2)
fig, axes = plt.subplots(ncols=2, nrows=2)
ax = axes.ravel()
ax[0].set_title('Original image')
ax[0].imshow(img)
ax[1].set_title('Mask')
ax[1].imshow(cvmask, cmap=plt.cm.gray)
#ax[2].set_title('Defected image')
#ax[2].imshow(image_defect)
def get_normal(depth_refine, fx=-1, fy=-1, cx=-1, cy=-1, for_vis=True):
res_y = depth_refine.shape[0]
res_x = depth_refine.shape[1]
# inpainting
scaleOri = np.amax(depth_refine)
inPaiMa = np.where(depth_refine == 0.0, 255, 0)
inPaiMa = inPaiMa.astype(np.uint8)
inPaiDia = 5.0
depth_refine = depth_refine.astype(np.float32)
depPaint = cv2.inpaint(depth_refine, inPaiMa, inPaiDia, cv2.INPAINT_NS)
depNorm = depPaint - np.amin(depPaint)
rangeD = np.amax(depNorm)
depNorm = np.divide(depNorm, rangeD)
depth_refine = np.multiply(depNorm, scaleOri)
centerX = cx
centerY = cy
constant = 1 / fx
uv_table = np.zeros((res_y, res_x, 2), dtype=np.int16)
column = np.arange(0, res_y)
uv_table[:, :, 1] = np.arange(0, res_x) - centerX # x-c_x (u)
uv_table[:, :, 0] = column[:, np.newaxis] - centerY # y-c_y (v)
uv_table_sign = np.copy(uv_table)
uv_table = np.abs(uv_table)
# kernel = np.ones((5, 5), np.uint8)
# depth_refine = cv2.dilate(depth_refine, kernel, iterations=1)
# depth_refine = cv2.medianBlur(depth_refine, 5 )
depth_refine = ndimage.gaussian_filter(depth_refine, 2) # sigma=3)
# depth_refine = ndimage.uniform_filter(depth_refine, size=11)
# very_blurred = ndimage.gaussian_filter(face, sigma=5)
v_x = np.zeros((res_y, res_x, 3))
v_y = np.zeros((res_y, res_x, 3))
normals = np.zeros((res_y, res_x, 3))
dig = np.gradient(depth_refine, 2, edge_order=2)
v_y[:, :, 0] = uv_table_sign[:, :, 1] * constant * dig[0]
v_y[:, :, 1] = depth_refine * constant + (uv_table_sign[:, :, 0] * constant) * dig[0]
v_y[:, :, 2] = dig[0]
v_x[:, :, 0] = depth_refine * constant + uv_table_sign[:, :, 1] * constant * dig[1]
v_x[:, :, 1] = uv_table_sign[:, :, 0] * constant * dig[1]
v_x[:, :, 2] = dig[1]
cross = np.cross(v_x.reshape(-1, 3), v_y.reshape(-1, 3))
norm = np.expand_dims(np.linalg.norm(cross, axis=1), axis=1)
# norm[norm == 0] = 1
cross = cross / norm
cross = cross.reshape(res_y, res_x, 3)
cross = np.abs(cross)
cross = np.nan_to_num(cross)
# cross_ref = np.copy(cross)
# cross[cross_ref==[0,0,0]]=0 #set zero for nan values
# cam_angle = np.arccos(cross[:, :, 2])
# cross[np.abs(cam_angle) > math.radians(75)] = 0 # high normal cut
cross[depth_refine <= 400] = 0 # 0 and near range cut
cross[depth_refine > depthCut] = 0 # far range cut
if not for_vis:
scaDep = 1.0 / np.nanmax(depth_refine)
depth_refine = np.multiply(depth_refine, scaDep)
cross[:, :, 0] = cross[:, :, 0] * (1 - (depth_refine - 0.5)) # nearer has higher intensity
cross[:, :, 1] = cross[:, :, 1] * (1 - (depth_refine - 0.5))
cross[:, :, 2] = cross[:, :, 2] * (1 - (depth_refine - 0.5))
scaCro = 255.0 / np.nanmax(cross)
cross = np.multiply(cross, scaCro)
cross = cross.astype(np.uint8)
return cross, depth_refine
def FindRemoveGlint(roi, cfg):
'''
Locate small bright region roughly centered in ROI
This function should be called before any major preprocessing of the frame.
The ROI should be unscaled and unblurred, since we assume glints
are saturated and have maximum intensity (255 in uint8)
Arguments
----
roi : 2D numpy uint8 array
Pupil/iris ROI image
cfg : configuration object
Configuration parameters including fractional glint diameter estimate
pupil_bw : 2D numpy unit8 array
Black and white pupil segmentation
Returns
----
glint : float array
N x 2 array of glint centroids
glint_mask : 2D numpy uint8 array
Bright region mask used by glint removal
roi_noglint : 2D numpy uint8 array
Pupil/iris ROI without small bright areas
'''
DEBUG = False
# ROI dimensions and center
ny, nx = roi.shape
roi_cx, roi_cy = nx / 2.0, ny / 2.0
if DEBUG:
print("%s, %s" % (roi_cx, roi_cy))
# Estimated glint diameter in pixels
glint_d = int(cfg.getfloat('PUPILSEG', 'glintdiameterperc') * nx / 100.0)
# Glint diameter should be >= 1 pixel
if glint_d < 1:
glint_d = 1
# Reasonable upper and lower bounds on glint area (x3, /3)
glint_A = np.pi * (glint_d / 2.0)**2
A_min, A_max = glint_A / 3.0, glint_A * 9.0
# print
# print A_min
# print A_max
# Find bright pixels in full scale uint8 image (ie value > 250)
bright = np.uint8(roi > 254)
# Label connected regions (blobs)
bright_labels = measure.label(bright, background=0) + 1
# Get region properties for all bright blobs in mask
bright_props = measure.regionprops(bright_labels)
# Init glint parameters
r_min = np.Inf
glint_label = -1
glint = (0, 0)
glint_mask = np.zeros_like(roi, dtype="uint8")
roi_noglint = roi.copy()
# Find closest blob to ROI center within glint area range
for props in bright_props:
# Blob area in pixels^2
A = props.area
# Only accept blobs with area in glint range
if A > A_min and A < A_max:
# Check distance from ROI center
cy, cx = props.centroid # (row, col)
r = np.sqrt((cx - roi_cx)**2 + (cy - roi_cy)**2)
if r < r_min:
r_min = r
glint_label = props.label
glint = (cx, cy)
if glint_label > 0:
# Construct glint mask
glint_mask = np.uint8(bright_labels == glint_label)
# Dilate glint mask
k = glint_d
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (k, k))
glint_mask = cv2.morphologyEx(glint_mask, cv2.MORPH_DILATE, kernel)
# Inpaint dilated glint in ROI
roi_noglint = cv2.inpaint(roi, glint_mask, 3, cv2.INPAINT_TELEA)
return glint, glint_mask, roi_noglint
# 2、扩张修复区域：
# 识别到修复区域并根据相邻像素值进行扩张达到弥补像素值修复图片的效果。cv2.inpaint（）函数主要涉及两种算法。
# 一种算法是从该区域的边界开始，然后进入区域内，逐渐填充边界中的所有内容。
# 它需要在邻近的像素周围的一个小邻域进行修复。该像素由邻居中所有已知像素的归一化加权和代替。
# 选择权重是一个重要的问题。对于靠近该点的那些像素，靠近边界的法线和位于边界轮廓上的像素，给予更多的权重。
# 另一种是基于流体动力学并利用偏微分方程。基本原则是heurisitic。
# 它首先沿着已知区域的边缘行进到未知区域（因为边缘是连续的）。
# 它继续等照片（连接具有相同强度的点的线，就像轮廓连接具有相同高度的点一样），同时在修复区域的边界处匹配渐变矢量。
# 为此，使用来自流体动力学的一些方法。获得颜色后，填充颜色以减少该区域的最小差异。
#扩张待修复区域
hi_mask = cv2.dilate(thresh, kernel, iterations=1)
specular = cv2.inpaint(img, hi_mask, 5, flags=cv2.INPAINT_TELEA)
cv2.namedWindow("Image", 0)
cv2.resizeWindow("Image", int(width / 2), int(hight / 2))
cv2.imshow("Image", img)
cv2.namedWindow("newImage", 0)
cv2.resizeWindow("newImage", int(width / 2), int(hight / 2))
a=cv2.imshow("newImage", specular)
cv2.imwrite("43.jpg",specular)
cv2.waitKey(0)
cv2.destroyAllWindows()
# 修复程序处理二的搭建
# 1、图像处理第二步：
def haircut(img_thresh, img, coordidate, sideLength):
coordidate_length = len(coordidate)
x = []
y = []
for i in range(coordidate_length):
coordidate_tuple = coordidate[i]
x_mid = int(coordidate_tuple[0] + coordidate_tuple[2]) // 2
y_mid = int(coordidate_tuple[1] + coordidate_tuple[3]) // 2
x.append(y_mid)
y.append(x_mid)
cv2.imshow('src', img)
cv2.waitKey(0)
rows = img_thresh.shape[0]
cols = img_thresh.shape[1]
# print(rows)
# print(cols)
for i in range(rows):
for j in range(cols):
if img_thresh[i][j] < 180:
img_thresh[i][j] = 0
else:
img_thresh[i][j] = 255
length = len(x)
# 对边界的点进行移动
for i in range(length):
if x[i] - sideLength // 2 < 0:
abs = sideLength // 2 - x[i]
x[i] = x[i] + abs
if x[i] + sideLength // 2 > rows:
abs = x[i] + sideLength / 2 - rows
x[i] = x[i] - abs
if y[i] - sideLength // 2 < 0:
abs = sideLength // 2 - y[i]
y[i] = y[i] + abs
if y[i] + sideLength // 2 > cols:
abs = y[i] + sideLength // 2 - cols
y[i] = y[i] - abs
x1 = []
y1 = []
x2 = []
y2 = []
for i in range(length):
a1 = int(x[i] - sideLength // 2)
b1 = int(y[i] - sideLength // 2)
a2 = int(x[i] + sideLength // 2)
b2 = int(y[i] + sideLength // 2)
x1.append(a1)
y1.append(b1)
x2.append(a2)
y2.append(b2)
# print(x1)
# print(y1)
# print(x2)
# print(y2)
#进行剃发的相关代码
# Convert the original image to grayscale
grayScale = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
# cv2.imwrite('grayScale_sample1.jpg', grayScale, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
# Kernel for the morphological filtering
kernel = cv2.getStructuringElement(1, (60, 60))
# Perform the blackHat filtering on the grayscale image to find the
# hair countours
blackhat = cv2.morphologyEx(grayScale, cv2.MORPH_BLACKHAT, kernel)
# cv2.imwrite('blackhat_sample1.jpg', blackhat, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
# intensify the hair countours in preparation for the inpainting
# algorithm
ret, thresh2 = cv2.threshold(blackhat, 10, 255, cv2.THRESH_BINARY)
print(thresh2.shape)
# cv2.imwrite('thresholded_sample1.jpg', thresh2, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
# inpaint the original image depending on the mask
dst = cv2.inpaint(img, thresh2, 1, cv2.INPAINT_TELEA)
cv2.imwrite('./pic/temp/background.jpg', dst,
[int(cv2.IMWRITE_JPEG_QUALITY), 90])
background = cv2.imread("./pic/temp/background.jpg")
#对光头图片进行毛发渲染
result = img.copy()
for i in range(length):
for m in range(x1[i], x2[i]):
for n in range(y1[i], y2[i]):
result[m][n] = [6, 6, 6]
print('完成')
for i in range(rows):
for j in range(cols):
if all(result[i][j] != [6, 6, 6]):
result[i][j] = background[i][j]
for i in range(length):
for m in range(x1[i], x2[i]):
for n in range(y1[i], y2[i]):
result[m][n] = img[m][n]
return result
def colour(counter, marker, distance):
if Timing:
Start_of_code = time.time()
results = [] # empty list
for k in range(1, counter + 1):
num_storage = [] # store tuple colour's value in RGB
name_storage = [] # stores the colour's name
img = cv2.imread("colour%d.png" % k)
height, width, numchannels = img.shape
roi = img[int((height / 2) -
(height / 2) * 0.85):int((height / 2) +
(height / 2) * 0.85),
int((width / 2) -
(width / 2) * 0.85):int((width / 2) +
(width / 2) * 0.85)]
nroi = cv2.resize(roi, (100, 100))
imgGray = cv2.cvtColor(nroi, cv2.COLOR_BGR2GRAY)
# img = cv2.imread("colour%d.png" % k, cv2.COLOR_BGR2RGB)
Gauss = cv2.GaussianBlur(imgGray, (5, 5), 0)
# blur = cv2.medianBlur(Gauss, 7)
# fliter = cv2.bilateralFilter(blur, 15, 75, 75)
# kernel = np.ones((10, 10), np.uint8)
# erode = cv2.erode(Gauss, kernel, iterations=10)
# dilation = cv2.dilate(erode, kernel, iterations=20)
# denoised = cv2.fastNlMeansDenoisingColored(dilation, None, 10, 10, 7, 21)
ret, otsu = cv2.threshold(Gauss, 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)
inpaint = cv2.inpaint(nroi, otsu, 3, cv2.INPAINT_NS)
h = 8
w = 8
# h, w = img[:2]
current_mode = 0
# # defined boundaries HSV
# boundaries = [("black", [0, 0, 0], [179, 255, 50]), ("white", [0, 0, 185], [179, 30, 255]),
# ("orange", [10, 30, 50], [25, 255, 255]), ("yellow", [25, 30, 50], [35, 255, 255]),
# ("green", [35, 30, 50], [85, 255, 255]), ("blue", [85, 30, 50], [130, 255, 255]),
# ("purple", [130, 30, 50], [145, 255, 255]), ("pink", [145, 30, 50], [165, 255, 255]),
# ("red", [165, 30, 50], [179, 255, 255]), ("red", [0, 30, 50], [10, 255, 255]),
# ("grey", [0, 0, 50], [179, 30, 185])]
# # defined boundaries RGB
# boundaries = [("black", [0, 0, 0]), ("white", [255, 255, 255]),
# ("orange", [255, 165, 0]), ("yellow", [255, 255, 0]),
# ("green", [0, 128, 0]), ("blue", [0, 0, 255]),
# ("purple", [128, 0, 128]), ("pink", [255, 192, 203]),
# ("red", [255, 0, 0]), ("grey", [128, 128, 128]),
# ("aqua", [0, 255, 255]), ("fuchsia", [255, 0, 255]),
# ("silver", [192, 192, 192]), ("maroon", [128, 0, 0]),
# ("olive", [128, 128, 0]), ("lime", [0, 255, 0]),
# ("teal", [0, 128, 128]), ("navy", [0, 0, 128]),]
# # defined boundaries RGB
# boundaries = [("black", [0, 0, 0]), ("white", [255, 255, 255]),
# ("yellow", [255, 255, 0]), ("purple", [128, 0, 128]),
# ("green", [0, 128, 0]), ("blue", [0, 0, 255]),
# ("red", [255, 0, 0]), ("grey", [128, 128, 128]),
# ("blue", [0, 255, 255]), ("pink", [255, 0, 255]),
# ("grey", [192, 192, 192]), ("red", [128, 0, 0]),
# ("yellow", [128, 128, 0]), ("green", [0, 255, 0]),
# ("blue", [0, 128, 128]), ("blue", [0, 0, 128])]
# # defined boundaries HSV
# boundaries = [("black", [0, 0, 0]), ("white", [0, 0, 255]),
# ("yellow", [30, 255, 255]), ("purple", [150, 255, 127]),
# ("green", [60, 255, 127]), ("blue", [120, 255, 255]),
# ("red", [0, 255, 255]), ("grey", [0, 0, 127]),
# ("blue", [90, 255, 255]), ("pink", [150, 255, 255]),
# ("grey", [0, 0, 191]), ("red", [0, 255, 127]),
# ("yellow", [30, 255, 127]), ("green", [60, 255, 255]),
# ("blue", [90, 255, 127]), ("blue", [120, 255, 127])]
# # defined boundaries HSV
# boundaries = [("black", [0, 0, 0], [179, 255, 50]), ("white", [0, 0, 179], [179, 38, 255]),
# ("orange", [15, 38, 50], [22, 255, 255]), ("yellow", [23, 38, 50], [44, 255, 255]),
# ("yellow green", [45, 38, 50], [52, 255, 255]), ("green", [53, 38, 50], [74, 255, 255]),
# ("green cyan", [75, 38, 50], [82, 255, 255]), ("cyan", [83, 38, 50], [104, 255, 255]),
# ("blue cyan", [105, 38, 50], [112, 255, 255]), ("blue", [113, 38, 50], [134, 255, 255]),
# ("violet", [135, 38, 50], [142, 255, 255]), ("magenta", [143, 38, 50], [164, 255, 255]),
# ("red magenta", [165, 38, 50], [172, 255, 255]), ("red", [0, 38, 50], [14, 255, 255]),
# ("red", [173, 38, 50], [180, 255, 255]), ("gray", [0, 0, 50], [179, 38, 179])]
# Colour Boundaries red, blue, yellow and gray HSV
# this requires dtype="uint16" for lower and upper & HSV = np.float32(HSV) before the conversion of HSV_FULL
boundaries = [("black", [0, 0, 0], [360, 255, 50]),
("white", [0, 0, 179], [360, 38, 255]),
("orange", [15, 38, 50], [31, 255, 255]),
("yellow", [31, 38, 50], [75, 255, 255]),
("yellow green", [75, 38, 50], [91, 255, 255]),
("green", [91, 38, 50], [135, 255, 255]),
("green cyan", [135, 38, 50], [150, 255, 255]),
("cyan", [150, 38, 50], [195, 255, 255]),
("blue cyan", [195, 38, 50], [210, 255, 255]),
("blue", [210, 38, 50], [255, 255, 255]),
("violet", [255, 38, 50], [270, 255, 255]),
("magenta", [270, 38, 50], [315, 255, 255]),
("red magenta", [315, 38, 50], [330, 255, 255]),
("red", [0, 38, 50], [15, 255, 255]),
("red", [330, 38, 50], [360, 255, 255]),
("gray", [0, 0, 50], [360, 38, 179])]
resizeBGR = cv2.resize(
nroi,
(w,
h)) # reduces the size of the image so the process would run fast
if Save_Data:
if not os.path.exists("color"):
os.makedirs("color")
color_result = "color/{0}_{1}.png".format(marker, k)
color_result_des = os.path.join(script_dir, color_result)
cv2.imwrite(color_result_des, nroi)
if Static_Test:
color_result = "{0}/{1}_{2}.png".format(distance, marker, k)
color_result_des = os.path.join(script_dir, color_result)
cv2.imwrite(color_result_des, nroi)
if Rover_Marker:
marker_name = "marker={0}".format(marker)
color_result = "{0}/{1}_{2}.png".format(marker_name, marker, k)
color_result_des = os.path.join(script_dir, color_result)
cv2.imwrite(color_result_des, nroi)
# print(resizeBGR[1,1])
resizeBGR = np.float32(resizeBGR)
resizeHSV = cv2.cvtColor(resizeBGR, cv2.COLOR_BGR2HSV_FULL)
resizeHSV[:, :, 1] = np.dot(resizeHSV[:, :, 1], 255)
# resizeRGB = cv2.cvtColor(resizeBGR, cv2.COLOR_BGR2RGB)
# print(resizeHSV[1,1])
if Step_color:
# for x in range(0, w):
# for y in range(0, h):
# num_storage.append(resizeRGB[x, y])
# print(num_storage)
cv2.imshow("fliter", dilation)
cv2.imshow("denos", denoised)
cv2.imshow("resize", resizeBGR)
cv2.waitKey(0)
# end if
# for i in range(0, h):
# for j in range(0, w):
# RGB = resizeHSV[i, j]
# differences = []
# for (name, value) in boundaries:
# for component1, component2 in zip(RGB, value):
# difference = sum([abs(component1 - component2)])
# differences.append([difference, name])
# differences.sort()
# name_storage.append(differences[0][1])
#
# majority = Counter(name_storage)
# results.append(majority.most_common(1)[0][0])
# for i in range(0, h):
# for j in range(0, w):
# # RGB = []
# RGB = resizeRGB[i, j]
# # num_storage.append(RGB)
# # Finds the nearest colour name within the webcolors dataset by converting the classification to rgb then then find the closest the square is to remove the negative value.
# try:
# colorname = webcolors.rgb_to_name(RGB)
# name_storage.append(colorname)
#
# except ValueError:
# min_colours = {}
# for key, name in webcolors.CSS3_HEX_TO_NAMES.items():
# r_c, g_c, b_c = webcolors.hex_to_rgb(key)
# rd = (r_c - RGB[0]) ** 2
# gd = (g_c - RGB[1]) ** 2
# bd = (b_c - RGB[2]) ** 2
# min_colours[(rd + gd + bd)] = name
# name_storage.append(min_colours[min(min_colours.keys())])
#
# majority = Counter(name_storage)
#
# results.append(majority.most_common(1)[0][0])
# comparing each pixel of the picture and append the colour name in to a list (BGR to RGB to get the name)
for (color, lower, upper) in boundaries:
lower = np.array(lower, dtype="uint16")
upper = np.array(upper, dtype="uint16")
mask = cv2.inRange(resizeHSV, lower, upper)
ratio = np.round(
(cv2.countNonZero(mask) / (resizeHSV.size / 3)) * 100, 2)
if ratio > current_mode:
current_mode = ratio
name_storage.append(color)
else:
pass
results.append(name_storage[-1])
mode = Counter(results)
if Step_color:
print(name_storage)
print(majority)
print(mode)
if mode == Counter():
colourname = "None"
else:
colourname = mode.most_common(1)[0][0]
if Timing:
Duration_of_color_recognition = time.time() - Start_of_code
print("Duraction of Colour Recognition = {0}".format(
Duration_of_color_recognition))
# with open('Character Color.csv', 'a') as csvfile: # for testing purposes
# filewriter = csv.writer(csvfile, delimiter=',', quotechar='|', quoting=csv.QUOTE_MINIMAL, lineterminator='\n')
# filewriter.writerow(
# [str(marker), str(Duration_of_color_recognition)])
return colourname
print('maxW = %d minW = %d maxH = %d minH = %d'%(max(widthList),min(widthList),max(heightList),min(heightList)))
print('Resize operation finished!')
##图像去除标记红框代码
for imgName in tqdm(os.listdir(dsr2)):
imgDir = dsr2+imgName
img = cv2.imread(imgDir)
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
h,w,c = img.shape
mask = np.zeros((h,w,c), np.uint8)
for i in h:
for j in w:
data = img[i][j]
#uint8做减法会溢出，需转类型到int后再做减法
if(int(data[0]) > 100 and abs(int(data[0])-int(data[1]))>50):
#创建impainting模板，即图像中红色的区域，mask只需要单通道
mask[i][j] +=255
dstimg = cv2.inpaint(img,mask[:,:,0],20,cv2.INPAINT_TELEA)
cv2.imwrite(dsr2+imgName,dstimg)
print('Get rid of the red box is finished!')
def warp_image(self, inputs):
image = np.array(inputs['left_image'])
if self.background_mode == 'background':
background_image = np.array(inputs['background'])
warped_image = np.zeros_like(image).astype(float)
warped_image = np.stack([warped_image] * 2, 0)
pix_locations = self.xs - inputs['disparity']
mask = self.get_occlusion_mask(pix_locations)
masked_pix_locations = pix_locations * mask - self.process_width * (
1 - mask)
# do projection - linear interpolate up to 1 pixel away
weights = np.ones((2, self.crop_height, self.process_width)) * 10000
for col in range(self.process_width - 1, -1, -1):
loc = masked_pix_locations[:, col]
loc_up = np.ceil(loc).astype(int)
loc_down = np.floor(loc).astype(int)
weight_up = loc_up - loc
weight_down = 1 - weight_up
mask = loc_up >= 0
mask[mask] = \
weights[0, np.arange(self.crop_height)[mask], loc_up[mask]] > weight_up[mask]
weights[0, np.arange(self.crop_height)[mask], loc_up[mask]] = \
weight_up[mask]
warped_image[0, np.arange(self.crop_height)[mask], loc_up[mask]] = \
image[:, col][mask] / 255.
mask = loc_down >= 0
mask[mask] = \
weights[1, np.arange(self.crop_height)[mask], loc_down[mask]] > weight_down[mask]
weights[1, np.arange(self.crop_height)[mask],
loc_down[mask]] = weight_down[mask]
warped_image[1, np.arange(self.crop_height)[mask], loc_down[mask]] = \
image[:, col][mask] / 255.
weights /= weights.sum(0, keepdims=True) + 1e-7 # normalise
weights = np.expand_dims(weights, -1)
warped_image = warped_image[0] * weights[1] + warped_image[
1] * weights[0]
warped_image *= 255.
# now fill occluded regions with random background
if self.background_mode == 'background':
warped_image[warped_image.max(-1) == 0] = background_image[
warped_image.max(-1) == 0]
warped_image = warped_image.astype(np.uint8)
else:
_h, _w = warped_image.shape[:2]
inpaint_mask = np.zeros(shape=(_h, _w, 1), dtype=np.uint8)
cmask = (warped_image.max(-1) == 0)
inpaint_mask[cmask] = 1
inputs['mask'] = inpaint_mask
if self.background_mode == 'opencv':
warped_image[cmask] = 0
warped_image = warped_image.astype(np.uint8)
warped_image = cv2.inpaint(warped_image, inpaint_mask, 3,
cv2.INPAINT_TELEA)
else:
warped_image[cmask] = 128
warped_image = warped_image.astype(np.uint8)
inputs['mask'] = inputs['mask'][:, :self.crop_width]
## Do this outside
# inference = self.Inpainter.inference(warped_image, inpaint_mask)
# inference = (tensor2img(inference) * 255).astype(int)
# warped_image = np.where(inpaint_mask == 0, warped_image, inference)
# plt.imshow(warped_image)
# plt.show()
# exit()
return warped_image
def get_negatives_stitching(self, path, mode=1):
import cv2
import os
import scipy.ndimage.morphology as morph
import random
# decide if the negative is of color or shape (only one, so that they cannot mix)
color_or_shape = np.random.randint(
0, 3) # 0 is color, 1 is shape, 2 is material
path_image_original = self.path_dataset + '/images/' + self.split + '/positive/' + path + '.' + self.image_extension
index_object_original = np.random.randint(0, 3)
image_original = cv2.imread(path_image_original)
seg_original = cv2.imread(
path_image_original.replace('positive', 'segmentation'))
with open(
os.path.join(self.path_dataset, 'scenes', self.split,
f'{path}.json'), 'rb') as f:
scene_original = json.load(f)
color_original = scene_original['objects'][index_object_original][
'color']
shape_original = scene_original['objects'][index_object_original][
'shape']
material_original = scene_original['objects'][index_object_original][
'material']
position = scene_original['objects'][index_object_original][
'pixel_coords'][:2]
# inpaint hole of previous object
mask_inpaint = (seg_original[..., 2] == index_object_original +
1).astype(np.uint8)
mask_inpaint = morph.binary_dilation(mask_inpaint,
structure=np.ones(
(5, 5))).astype(np.uint8)
image_inpainted = cv2.inpaint(image_original, mask_inpaint, 3,
cv2.INPAINT_TELEA)
# open other image
other_correct = False
while not other_correct:
path_other = random.choice(list(self.paths.values()))
index_object_other = np.random.randint(0, 3)
with open(
os.path.join(self.path_dataset, 'scenes', self.split,
f'{path_other}.json'), 'rb') as f:
scene_other = json.load(f)
color_other = scene_other['objects'][index_object_other]['color']
shape_other = scene_other['objects'][index_object_other]['shape']
material_other = scene_other['objects'][index_object_other][
'material']
# brute force...
if mode == 1:
if color_or_shape == 1 and (color_other != color_original or
material_other != material_original
or shape_other == shape_original):
continue
elif color_or_shape == 0 and (
color_other == color_original
or material_other != material_original
or shape_other != shape_original):
continue
elif color_or_shape == 2 and (
color_other != color_original or material_other
== material_original or shape_other != shape_original):
continue
else:
other_correct = True
elif mode == 2:
if color_or_shape == 1 and (shape_other == shape_original):
continue
elif color_or_shape == 0 and (color_other == color_original):
continue
elif color_or_shape == 2 and (material_other
== material_original):
continue
else:
other_correct = True
path_image_other = self.path_dataset + '/images/' + self.split + '/positive/' + path_other + '.' + self.image_extension
image_other = cv2.imread(path_image_other)
seg_other = cv2.imread(
path_image_other.replace('positive', 'segmentation'))
# stitch object
# %10 in the case of SCLEVR3, if not, it does not harm
mask_stitch = (seg_other[..., 2] % 10) == index_object_other + 1
mask_stitch = mask_stitch.astype(np.uint8)
# extend mask so that the gradients (borders) are visible
mask_stitch = morph.binary_dilation(mask_stitch,
structure=np.ones(
(10, 10))).astype(np.uint8)
mask_stitch = np.stack((mask_stitch, ) * 3, -1) * 255
try:
image_stitched = cv2.seamlessClone(image_other, image_inpainted,
mask_stitch, tuple(position),
cv2.NORMAL_CLONE)
except:
print('retry')
return self.get_negatives_stitching(path)
img = Image.fromarray(image_stitched).convert('RGB')
img = self.transform(img)
return img
def extract(anns: list, img: np.ndarray, config: AugSegConfig):
"""
Inpaint background, extract list of instances (in overlapping groups), get transformations and other useful
information from input image and annotations.
:param anns: list of input annotations
:param img: input original image
:param config: an AugSegConfig instance containing transform parameters
:return: [background: inpainted background image
instances_list: list of instance rgba images
transforms_list: list of transformation dicts
groupbnd_list: list of bounding boxes
group: list of instence indices in groups]
"""
width = img.shape[1]
height = img.shape[0]
mask, labels = __get_coco_masks(anns, height, width)
# inpainting
#inpaint_mask = np.uint8(mask > 0)
#inpaint_mask = ndimage.binary_dilation(inpaint_mask, structure=ndimage.generate_binary_structure(2, 2), iterations=2)
background = cv2.inpaint(img, np.uint8(mask), 5, cv2.INPAINT_NS)
numinst = len(anns)
bboxs = [ann['bbox'] for ann in anns]
inst_group_belonging = [0] * numinst
group = []
group_idx = 1
for i in range(numinst):
if inst_group_belonging[i] == 0:
group.append([i])
inst_group_belonging[i] = group_idx
dfs(bboxs, inst_group_belonging, group[len(group) - 1], group_idx, i)
group_idx += 1
realbboxs = []
instances_list = []
transforms_list = []
groupbnd_list = []
for i in range(len(group)):
x, y, w, h = bboxs[group[i][0]]
realbboxs.append([x, y, x + w, y + h])
for j in range(len(group[i])):
x, y, w, h = bboxs[group[i][j]]
realbboxs[i][0] = min(realbboxs[i][0], x)
realbboxs[i][1] = min(realbboxs[i][1], y)
realbboxs[i][2] = max(realbboxs[i][2], x + w)
realbboxs[i][3] = max(realbboxs[i][3], y + h)
xmin, ymin, xmax, ymax = realbboxs[i]
maskgroupi = get_masks(mask, group[i])
trimapi = gettrimap(maskgroupi, 5)
alphamapi = global_matting(img, trimapi)
alphamapi = guided_filter(img, trimapi, alphamapi, 10, 1e-5)
ymin, ymax, xmin, xmax = [int(round(x)) for x in (ymin, ymax, xmin, xmax)]
resulti = np.dstack((img[ymin:ymax, xmin:xmax], alphamapi[ymin:ymax, xmin:xmax]))
restricts = get_restriction([xmin, ymin, xmax, ymax], width, height)
resulti, transformi = get_transform(resulti, restricts, config) # resulti may be flipped
transformi['tx'] += (xmin + xmax) / 2
transformi['ty'] += (ymin + ymax) / 2
instances_list.append(resulti)
transforms_list.append(transformi)
groupbnd_list.append([xmin, ymin, xmax, ymax])
return background, instances_list, transforms_list, groupbnd_list, group
def inpaint(image, mask):
RADIOUS = 3
return cv2.inpaint(image, mask, RADIOUS, cv2.INPAINT_TELEA)
def __apply_inpaint(self, frame, mask):
output = cv.inpaint(frame, mask, 3, cv.INPAINT_TELEA)
return output
# -*- coding: utf-8 -*-
from PIL import Image
from googletrans import Translator
import pytesseract
import os
import cv2
import imutils
path= 'C:/Users/yyw/Downloads/taobao/a/TB24c7ls_lYBeNjSszcXXbwhFXa_!!2871484755.jpg'
img = cv2.imread(path)
mask = cv2.threshold(img, 210, 255, cv2.THRESH_BINARY)[1][:,:,0]
dst = cv2.inpaint(img, mask, 7, cv2.INPAINT_NS)
cv2.imshow("input image",dst)
cv2.waitKey(0)
upper2 = np.array([7, 255, 255], dtype=np.uint8)
mask2 = cv2.inRange(hsvImage, lower2, upper2)
mask = cv2.bitwise_or(mask1, mask2)
# Perform morphological close operation to connect the laser lines
kernel = np.ones((5, 5), np.uint8)
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel, iterations=15)
# Subtract red lazer beams from the image
channels = cv2.split(image)
maskDilation = cv2.dilate(mask, kernel, iterations=5)
maskInverse = cv2.bitwise_not(maskDilation)
redChannel = cv2.bitwise_and(channels[2], maskInverse)
image = cv2.merge([channels[0], channels[1], redChannel])
image = cv2.inpaint(image, maskDilation, 3, cv2.INPAINT_TELEA)
# Find and draw contours
image2, contours, hierarchy = cv2.findContours(mask, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours:
if (cv2.contourArea(cnt) >= 10000 and cv2.contourArea(cnt) <= 3000000):
rect = cv2.minAreaRect(cnt)
box = cv2.boxPoints(rect)
box = np.int0(box)
cv2.drawContours(image, [box], 0, (255, 0, 0), 50)
# cv2.drawContours(image, contours, -1, (0,255,0), 5)
# Display the result
result = cv2.cvtColor(redChannel, cv2.COLOR_GRAY2BGR)
def main(argv):
# [load_image]
if len(argv) < 1:
print('Not enough parameters')
print('Usage:\nmorph_lines_detection.py < path_to_image >')
return -1
# print(argv[0])
argv = base64.b64decode(argv).decode("utf-8")
# Load the image
src = imread(argv)
# Check if image is loaded fine
if src is None:
print('Error opening image: ' + argv)
return -1
# Read the image
img = cv2.imread(argv, 0)
# Thresholding the image
(thresh, img_bin) = cv2.threshold(img, 128, 255,
cv2.THRESH_BINARY | cv2.THRESH_OTSU)
# Invert the image
img_bin = 255 - img_bin
# cv2.imwrite("C:/ICR/uploads/tempFileName_2019041809585560-01.jpg", img_bin)
# Defining a kernel length
kernel_length = np.array(img).shape[1] // 60
# A verticle kernel of (1 X kernel_length), which will detect all the verticle lines from the image.
verticle_kernel = cv2.getStructuringElement(cv2.MORPH_RECT,
(1, kernel_length))
# A horizontal kernel of (kernel_length X 1), which will help to detect all the horizontal line from the image.
hori_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (kernel_length, 1))
# A kernel of (3 X 3) ones.
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
# Morphological operation to detect vertical lines from an image
img_temp1 = cv2.erode(img_bin, verticle_kernel, iterations=2)
verticle_lines_img = cv2.dilate(img_temp1, verticle_kernel, iterations=2)
#cv2.imwrite("C:/ICR/uploads/verticle_lines.jpg", verticle_lines_img)
# Morphological operation to detect horizontal lines from an image
img_temp2 = cv2.erode(img_bin, hori_kernel, iterations=2)
horizontal_lines_img = cv2.dilate(img_temp2, hori_kernel, iterations=2)
# Weighting parameters, this will decide the quantity of an image to be added to make a new image.
alpha = 0.5
beta = 1.0 - alpha
# This function helps to add two image with specific weight parameter to get a third image as summation of two image.
img_final_bin = cv2.addWeighted(verticle_lines_img, alpha,
horizontal_lines_img, beta, 0.0)
img_final_bin = cv2.erode(~img_final_bin, kernel, iterations=2)
(thresh,
img_final_bin) = cv2.threshold(img_final_bin, 128, 255,
cv2.THRESH_BINARY | cv2.THRESH_OTSU)
# 테두리 저장
img_final_bin = 255 - img_final_bin
# merge
dst = cv2.inpaint(img, img_final_bin, 5, cv2.INPAINT_TELEA)
cv2.imwrite(argv, dst)
print(str(round(rate * 100)) + '%')
rate = rate + 0.1
kernel = np.ones((3, 3), np.uint8)
mask_list[i] = cv2.dilate(mask_list[i], kernel, iterations=1)
mask_list[i] = cv2.morphologyEx(mask_list[i], cv2.MORPH_OPEN, kernel)
del kernel
gc.collect()
# マスク領域を除去し周りの画素から補完
print('Image Inpainting...')
rate = 0.1
for i in range(fnum):
if ((i / fnum) >= rate):
print(str(round(rate * 100)) + '%')
rate = rate + 0.1
frame_list[i] = cv2.inpaint(frame_list[i], mask_list[i], 3,
cv2.INPAINT_TELEA)
del mask_list
gc.collect()
# TLDの初期化
print('Making Tracker...')
# 1人目の追跡器の追跡器
player1, b1 = None, None
player1, b1 = InitializingTLD(player1, b1, frame_list[0])
# 2人目の追跡器の初期化
player2, b2 = None, None
player2, b2 = InitializingTLD(player2, b2, frame_list[0])
# TLDの初期化
print('Tracking Learning Detection...')
rate = 0.1
import cv2
import numpy as np
# load gambar yang rusak
image = cv2.imread('abraham.jpg')
cv2.imshow('Original Damaged Photo', image)
cv2.waitKey(0)
# load gambar yang sudah di mark bagian yang rusak
marked_damages = cv2.imread('mask.jpg', 0)
cv2.imshow('Marked Damages', marked_damages)
cv2.waitKey(0)
# kasih threshold biar keliatan jelas bagian yang rusak
ret, thresh1 = cv2.threshold(marked_damages, 254, 255, cv2.THRESH_BINARY)
cv2.imshow('Threshold Binary', thresh1)
cv2.waitKey(0)
# gambar di dilate biar bagian yang rusak terlihat makin lebar
kernel = np.ones((7, 7), np.uint8)
mask = cv2.dilate(thresh1, kernel, iterations=1)
cv2.imshow('Dilated Mask', mask)
cv2.imwrite('images/abraham_mask.png', mask)
cv2.waitKey(0)
# fungsi utama yang digunakan untuk photo restoration
restored = cv2.inpaint(image, mask, 3, cv2.INPAINT_TELEA)
cv2.imshow('Restored', restored)
cv2.waitKey(0)
cv2.destroyAllWindows()
def inpaint_depth(self, rad, depth=None):
"""inpaint depth"""
if depth is None: depth = self.depth
return cv2.inpaint(depth, self.invalid_depth_mask(depth=depth), rad, cv2.INPAINT_TELEA)
if __name__ == '__main__':
import sys
try:
fn = sys.argv[1]
except:
fn = '../data/fruits.jpg'
print(__doc__)
img = cv2.imread(fn)
if img is None:
print('Failed to load image file:', fn)
sys.exit(1)
img_mark = img.copy()
mark = np.zeros(img.shape[:2], np.uint8)
sketch = Sketcher('img', [img_mark, mark], lambda : ((255, 255, 255), 255))
while True:
ch = 0xFF & cv2.waitKey()
if ch == 27:
break
if ch == ord(' '):
res = cv2.inpaint(img_mark, mark, 3, cv2.INPAINT_TELEA)
cv2.imshow('inpaint', res)
if ch == ord('r'):
img_mark[:] = img
mark[:] = 0
sketch.show()
cv2.destroyAllWindows()
def convertToNumpy(inputPath, idx):
# copied from datasetVideo.py
inputExtension = ".exr"
def get_image_name(i, j, mode):
if mode == 'high':
return os.path.join(inputPath,
"high_tmp_%05d%s" % (j, inputExtension))
if mode == 'highdn':
return os.path.join(inputPath,
"high_tmp_%05d_depth%s" % (j, inputExtension))
if mode == 'highfx':
return os.path.join(inputPath,
"high_tmp_%05d_fx%s" % (j, inputExtension))
elif mode == 'low':
return os.path.join(inputPath,
"low_tmp_%05d%s" % (j, inputExtension))
elif mode == 'dn':
return os.path.join(inputPath,
"low_tmp_%05d_depth%s" % (j, inputExtension))
elif mode == 'flow':
return os.path.join(inputPath,
"low_tmp_%05d_flow%s" % (j, inputExtension))
high = [None] * numFrames
low = [None] * numFrames
flow = [None] * numFrames
for j in range(numFrames):
high_rgb = np.clip(
np.asarray(imageio.imread(get_image_name(idx, j,
'high'))).transpose(
(2, 0, 1)), 0, 1)
high_dn = np.asarray(imageio.imread(get_image_name(
idx, j, 'highdn'))).transpose((2, 0, 1))
high_fx = np.asarray(imageio.imread(get_image_name(
idx, j, 'highfx'))).transpose((2, 0, 1))
high[j] = np.concatenate(
(high_rgb[3:4, :, :], high_dn, high_fx[0:1, :, :]), axis=0)
high[j][0, :, :] = high[j][0, :, :] * 2 - 1
assert high[j].shape[0] == 6
low_rgb = np.clip(
np.asarray(imageio.imread(get_image_name(idx, j,
'low'))).transpose(
(2, 0, 1)), 0, 1)
low_dn = np.asarray(imageio.imread(get_image_name(idx, j,
'dn'))).transpose(
(2, 0, 1))
low[j] = np.concatenate((low_rgb[3:4], low_dn), axis=0)
low[j][0, :, :] = low[j][0, :, :] * 2 - 1 # transform mask to [-1,1]
assert low[j].shape[0] == 5
flow_xy = imageio.imread(get_image_name(idx, j, 'flow'))[:, :, 0:2]
flow_inpaint = np.stack(
(cv.inpaint(flow_xy[:, :, 0], np.uint8(low_rgb[3, :, :] == 0), 3,
cv.INPAINT_NS),
cv.inpaint(flow_xy[:, :, 1], np.uint8(low_rgb[3, :, :] == 0), 3,
cv.INPAINT_NS)),
axis=0)
flow[j] = flow_inpaint
images_high = np.stack(high, axis=0)
images_low = np.stack(low, axis=0)
flow_low = np.stack(flow, axis=0)
# save as numpy array
np.save(os.path.join(inputPath, "high_%05d.npy" % idx), images_high)
np.save(os.path.join(inputPath, "low_%05d.npy" % idx), images_low)
np.save(os.path.join(inputPath, "flow_%05d.npy" % idx), flow_low)