mattmcw
/
animation
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
animation/fourcell/normalize.py

465 lines
12 KiB
Python
Raw Blame History

import sys
import cv2
import numpy as np
import math
def image_resize(image, width = None, height = None, inter = cv2.INTER_AREA):
dim = None
(h, w) = image.shape[:2]
if width is None and height is None:
return image
if width is None:
r = height / float(h)
dim = (int(w * r), height)
else:
r = width / float(w)
dim = (width, int(h * r))
resized = cv2.resize(image, dim, interpolation = inter)
return resized
def display (image) :
resized = image_resize(image, 800, 800)
cv2.imshow('img', resized)
while cv2.getWindowProperty('img', cv2.WND_PROP_VISIBLE) > 0:
key = cv2.waitKey(0)
if key == 27:
cv2.destroyAllWindows()
break
exit(0)
def get_center (contour) :
M = cv2.moments(contour)
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
return cX, cY
def draw_line (image, hps, a, b) :
print(f'{a} -> {b}')
lA = (hps[a-1]['x'], hps[a-1]['y'])
lB = (hps[b-1]['x'], hps[b-1]['y'])
cv2.line(image, lA, lB, [0, 255, 0], 10)
return (lA, lB)
def horiz_angle (line, rotate = 0) :
deltaY = line[1][1] - line[0][1] #P2_y - P1_y
deltaX = line[1][0] - line[0][0] #P2_x - P1_x
angleInDegrees = normalize_angle(math.degrees(math.atan2(deltaY, deltaX) + rotate))
return angleInDegrees
def verts_angle (line) :
angleInDegrees = normalize_angle(horiz_angle(line, math.pi/2))
return angleInDegrees
def is_close (point, points) :
for pt in points :
if math.dist(point, pt) < 100 :
return True
return False
# taken from
# https://gist.github.com/phn/1111712/35e8883de01916f64f7f97da9434622000ac0390
def normalize_angle (num, lower=0.0, upper=360.0, b=False):
"""Normalize number to range [lower, upper) or [lower, upper].
Parameters
----------
num : float
The number to be normalized.
lower : float
Lower limit of range. Default is 0.0.
upper : float
Upper limit of range. Default is 360.0.
b : bool
Type of normalization. See notes.
Returns
-------
n : float
A number in the range [lower, upper) or [lower, upper].
Raises
------
ValueError
If lower >= upper.
Notes
-----
If the keyword `b == False`, the default, then the normalization
is done in the following way. Consider the numbers to be arranged
in a circle, with the lower and upper marks sitting on top of each
other. Moving past one limit, takes the number into the beginning
of the other end. For example, if range is [0 - 360), then 361
becomes 1. Negative numbers move from higher to lower
numbers. So, -1 normalized to [0 - 360) becomes 359.
If the keyword `b == True` then the given number is considered to
"bounce" between the two limits. So, -91 normalized to [-90, 90],
becomes -89, instead of 89. In this case the range is [lower,
upper]. This code is based on the function `fmt_delta` of `TPM`.
Range must be symmetric about 0 or lower == 0.
Examples
--------
>>> normalize(-270,-180,180)
90
>>> import math
>>> math.degrees(normalize(-2*math.pi,-math.pi,math.pi))
0.0
>>> normalize(181,-180,180)
-179
>>> normalize(-180,0,360)
180
>>> normalize(36,0,24)
12
>>> normalize(368.5,-180,180)
8.5
>>> normalize(-100, -90, 90, b=True)
-80.0
>>> normalize(100, -90, 90, b=True)
80.0
>>> normalize(181, -90, 90, b=True)
-1.0
>>> normalize(270, -90, 90, b=True)
-90.0
"""
# abs(num + upper) and abs(num - lower) are needed, instead of
# abs(num), since the lower and upper limits need not be 0. We need
# to add half size of the range, so that the final result is lower +
# <value> or upper - <value>, respectively.
res = num
if not b:
if lower >= upper:
raise ValueError("Invalid lower and upper limits: (%s, %s)" %
(lower, upper))
res = num
if num > upper or num == lower:
num = lower + abs(num + upper) % (abs(lower) + abs(upper))
if num < lower or num == upper:
num = upper - abs(num - lower) % (abs(lower) + abs(upper))
res = lower if res == upper else num
else:
total_length = abs(lower) + abs(upper)
if num < -total_length:
num += math.ceil(num / (-2 * total_length)) * 2 * total_length
if num > total_length:
num -= math.floor(num / (2 * total_length)) * 2 * total_length
if num > upper:
num = total_length - num
if num < lower:
num = -total_length - num
res = num * 1.0 # Make all numbers float, to be consistent
return res
def mean (lst):
return sum(lst) / len(lst)
def find_hole_punches (img) :
left=-1
right=-1
top=-1
bottom=-1
if orientation :
left = width * 0.2
right = width * 0.8
else :
top = height * 0.2
bottom = height * 0.8
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
blur = cv2.medianBlur(gray, 31)
ret, thresh = cv2.threshold(blur, 200, 255, cv2.THRESH_BINARY)
canny = cv2.Canny(thresh, 75, 200)
contours, hierarchy = cv2.findContours(canny, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
contourList = []
areaList = []
for contour in contours:
approx = cv2.approxPolyDP(contour, 0.03 * cv2.arcLength(contour, True), True)
if cv2.isContourConvex(approx) :
cX, cY = get_center(contour)
if (orientation and ( cX < left or cX > right) ) or ( not orientation and ( cY < top or cY > bottom)) :
area = cv2.contourArea(contour)
areaList.append(area)
contourList.append(contour)
maxArea=0
maxIndex=0
#reduce to lambda
for i in range(len(areaList)) :
area = areaList[i]
if area > maxArea:
maxArea = area
maxIndex = i
count = 0
holePunches = []
centers = []
areaRange = 0
topLeft = None
minDist = 1000000
# pretty good
# add position constraint
while count < 6 :
areaRange+=1
for i in range(len(areaList)) :
area = areaList[i]
if area == maxArea or area * ((100 + areaRange) / 100) > maxArea :
cX, cY = get_center(contourList[i])
if is_close((cX, cY), centers) :
continue
centers.append((cX, cY))
print(f'{cX},{cY}')
hp = {
'x' : cX,
'y' : cY,
'contour' : contourList[i],
'dist' : math.dist((cX, cY), (0, 0)),
'order': -1
}
if hp['dist'] < minDist :
minDist = hp['dist']
topLeft = hp
holePunches.append(hp)
count+=1
for hp in holePunches :
hp['dist'] = math.dist( (topLeft['x'], topLeft['y']), (hp['x'], hp['y']) )
print(f'Hole punches: {len(holePunches)}')
print(f'Found hole punches within {areaRange}% of largest')
if len(holePunches) != 6:
print(f'Wrong number of hole punches, exiting...')
exit(4)
holePunches = sorted(holePunches, key = lambda hp: hp['dist'])
i = 0
for hp in holePunches :
hp['order'] = i
#cv2.putText(img, str(i + 1), (hp['x'], hp['y']), cv2.FONT_HERSHEY_SIMPLEX, 20, (0, 0, 255), 5, cv2.LINE_AA, False)
i+=1
return holePunches
def correct_rotation (img, original, holePunches) :
horizLines = [
(3, 1),
(6, 4),
(5, 2)
]
vertsLines = [
(1, 2),
(1, 4), #double long left
(1, 4), #
(3, 5),
(3, 6), #double long right
(3, 6), #
(2, 4),
(5, 6)
]
rotations = []
for h in horizLines :
line = draw_line(img, holePunches, h[0], h[1])
angle = horiz_angle(line)
print(angle)
rotations.append(angle)
for v in vertsLines :
line = draw_line(img, holePunches, v[0], v[1])
angle = verts_angle(line)
print(angle)
rotations.append(angle)
correctionRotation = mean(rotations) - 180
print(f'Mean rotation: {correctionRotation}')
(cX, cY) = (width // 2, height // 2)
M = cv2.getRotationMatrix2D((cX, cY), correctionRotation, 1.0)
#create rotation of original
return cv2.warpAffine(original, M, (width, height))
def create_blank (w, h, rgb_color = (255, 255, 255)) :
blank = np.zeros([h, w, 3], dtype=np.uint8)
color = tuple(reversed(rgb_color))
blank[:] = color
return blank
def get_mean_rect (holePunches) :
left = 0
right = 0
top = 0
bottom = 0
for hp in holePunches :
if hp['order'] == 0 :
left += float(hp['x'])
top += float(hp['y'])
elif hp['order'] == 2 :
right += float(hp['x'])
top += float(hp['y'])
elif hp['order'] == 3 :
left += float(hp['x'])
bottom += float(hp['y'])
elif hp['order'] == 5 :
right += float(hp['x'])
bottom += float(hp['y'])
w = round((right / 2.0) - (left / 2.0))
h = round((bottom / 2.0) - (top / 2.0))
return (w, h)
def center_within (larger, smaller) :
w1 = larger[0]
h1 = larger[1]
w2 = smaller[0]
h2 = smaller[1]
x = ((w1 - w2) / 2)
y = ((h1 - h2) / 2)
return (int(x), int(y))
# If we consider (0,0) as top left corner of image called
# im with left-to-right as x direction and top-to-bottom
# as y direction. and we have (x1,y1) as the top-left vertex
# and (x2,y2) as the bottom-right vertex of a rectangle
# region within that image, then:
#
# roi = im[y1:y2, x1:x2]
def crop (img, xoffset, yoffset, w, h) :
#crop_img = img[y:y+h, x:x+w].copy()
return im[yoffset:yoffset+w, xoffset:xoffset+w].copy()
def normalize_image(blank, rotated, offset, tl) :
rotatedHeight, rotatedWidth = rotated.shape[:2]
normalHeight, width = blank.shape[:2]
diffX = offset[0] - tl["x"]
diffY = offset[1] - tl["y"]
print(f'diffX : {diffX}')
print(f'diffY : {diffY}')
crop = rotated.copy()
if diffX < 0 :
crop = crop[0:rotatedHeight, abs(diffX):rotatedWidth]
rotatedHeight, rotatedWidth = crop.shape[:2]
print('Cropped X')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
diffX = 0
if diffY < 0 :
crop = crop[abs(diffY):rotatedHeight, 0:rotatedWidth]
rotatedHeight, rotatedWidth = crop.shape[:2]
print('Cropped Y')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
diffY = 0
if rotatedWidth > width :
crop = crop[0:rotatedHeight, 0:rotatedWidth-(rotatedWidth - width)]
rotatedHeight, rotatedWidth = crop.shape[:2]
print('Cropped X')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
if rotatedHeight > normalHeight :
crop = crop[0:rotatedHeight-(rotatedHeight - normalHeight), 0:width]
rotatedHeight, rotatedWidth = crop.shape[:2]
print('Cropped Y')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
print(f'diffX : {diffX}')
print(f'diffY : {diffY}')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
print(f'Blank : {width},{normalHeight}')
cropHeight = normalHeight
cropWidth = width
if normalHeight > rotatedHeight :
cropHeight = rotatedHeight
if width > rotatedWidth :
cropWidth = rotatedWidth
blank[diffY:cropHeight, diffX:cropWidth] = crop[0:cropHeight-diffY, 0:cropWidth - diffX]
return blank
if len(sys.argv) < 2:
print('Please provide path of scan to normalize')
exit(1)
if len(sys.argv) < 3:
print('Please provide path to output file')
exit(2)
scanImage = sys.argv[-2]
normalImage = sys.argv[-1]
pageDim = (11, 8.5)
pageRatio = pageDim[1] / pageDim[0]
print(f'Normalizing {scanImage} as {normalImage}')
original = cv2.imread(scanImage)
img = original.copy()
height, width = img.shape[:2]
orientation = height > width
if not orientation :
print(f'Scan is not in portrait mode, exiting...')
exit(3)
normalHeight = round(float(width) / pageRatio)
holePunches = find_hole_punches(img)
rotated = correct_rotation(img, original, holePunches)
rotatedHeight, rotatedWidth = rotated.shape[:2]
holePunches = find_hole_punches(rotated)
blank = create_blank(width, normalHeight)
tl = None
for hp in holePunches :
if tl is None :
tl = hp
print(f'{hp["order"] + 1} {hp["x"]},{hp["y"]}')
print(f'Normal dimensions: {width},{normalHeight}')
# the mean rectangle is the average width and height
# determined by the four corner hole punches
meanRect = get_mean_rect(holePunches)
print(f'Mean rectangle: {meanRect[0]},{meanRect[1]}')
# offset is the position within the new normal image
# the top left hole punch should be centered to
offset = center_within((width, normalHeight), meanRect)
print(f'Offset : {offset[0]},{offset[1]}')
print(f'Topleft: {tl["x"]},{tl["y"]}')
print(f'Rotated: {rotatedWidth},{rotatedHeight}')
print(f'Blank : {width},{normalHeight}')
#cv2.rectangle(blank, offset, (offset[0]+meanRect[0], offset[1]+meanRect[1]), (255, 0, 0), thickness=20)
normal = normalize_image(blank, rotated, offset, tl)
print(f'Writing normalized image to {normalImage}')
cv2.imwrite(normalImage, normal)
evaluation = find_hole_punches(normal)
with open(f'{normalImage}.txt', 'w') as evalFile :
for hp in evaluation:
evalFile.write(f'{hp["order"] + 1} : {hp["x"]},{hp["y"]}\n')
#display(normal)
Reference in New Issue View Git Blame Copy Permalink