auto cropping now works with non square crops
This commit is contained in:
captin411
GitHub
parent
0ddaf8d202
commit
1be5933ba2
|
|
@ -1,241 +1,270 @@
|
||||||
import cv2
|
import cv2
|
||||||
from collections import defaultdict
|
from collections import defaultdict
|
||||||
from math import log, sqrt
|
from math import log, sqrt
|
||||||
import numpy as np
|
import numpy as np
|
||||||
from PIL import Image, ImageDraw
|
from PIL import Image, ImageDraw
|
||||||
|
|
||||||
GREEN = "#0F0"
|
GREEN = "#0F0"
|
||||||
BLUE = "#00F"
|
BLUE = "#00F"
|
||||||
RED = "#F00"
|
RED = "#F00"
|
||||||
|
|
||||||
|
|
||||||
def crop_image(im, settings):
|
def crop_image(im, settings):
|
||||||
""" Intelligently crop an image to the subject matter """
|
""" Intelligently crop an image to the subject matter """
|
||||||
if im.height > im.width:
|
|
||||||
im = im.resize((settings.crop_width, settings.crop_height * im.height // im.width))
|
scale_by = 1
|
||||||
elif im.width > im.height:
|
if is_landscape(im.width, im.height):
|
||||||
im = im.resize((settings.crop_width * im.width // im.height, settings.crop_height))
|
scale_by = settings.crop_height / im.height
|
||||||
else:
|
elif is_portrait(im.width, im.height):
|
||||||
im = im.resize((settings.crop_width, settings.crop_height))
|
scale_by = settings.crop_width / im.width
|
||||||
|
elif is_square(im.width, im.height):
|
||||||
if im.height == im.width:
|
if is_square(settings.crop_width, settings.crop_height):
|
||||||
return im
|
scale_by = settings.crop_width / im.width
|
||||||
|
elif is_landscape(settings.crop_width, settings.crop_height):
|
||||||
focus = focal_point(im, settings)
|
scale_by = settings.crop_width / im.width
|
||||||
|
elif is_portrait(settings.crop_width, settings.crop_height):
|
||||||
# take the focal point and turn it into crop coordinates that try to center over the focal
|
scale_by = settings.crop_height / im.height
|
||||||
# point but then get adjusted back into the frame
|
|
||||||
y_half = int(settings.crop_height / 2)
|
im = im.resize((int(im.width * scale_by), int(im.height * scale_by)))
|
||||||
x_half = int(settings.crop_width / 2)
|
|
||||||
|
if im.width == settings.crop_width and im.height == settings.crop_height:
|
||||||
x1 = focus.x - x_half
|
if settings.annotate_image:
|
||||||
if x1 < 0:
|
d = ImageDraw.Draw(im)
|
||||||
x1 = 0
|
rect = [0, 0, im.width, im.height]
|
||||||
elif x1 + settings.crop_width > im.width:
|
rect[2] -= 1
|
||||||
x1 = im.width - settings.crop_width
|
rect[3] -= 1
|
||||||
|
d.rectangle(rect, outline=GREEN)
|
||||||
y1 = focus.y - y_half
|
if settings.destop_view_image:
|
||||||
if y1 < 0:
|
im.show()
|
||||||
y1 = 0
|
return im
|
||||||
elif y1 + settings.crop_height > im.height:
|
|
||||||
y1 = im.height - settings.crop_height
|
focus = focal_point(im, settings)
|
||||||
|
|
||||||
x2 = x1 + settings.crop_width
|
# take the focal point and turn it into crop coordinates that try to center over the focal
|
||||||
y2 = y1 + settings.crop_height
|
# point but then get adjusted back into the frame
|
||||||
|
y_half = int(settings.crop_height / 2)
|
||||||
crop = [x1, y1, x2, y2]
|
x_half = int(settings.crop_width / 2)
|
||||||
|
|
||||||
if settings.annotate_image:
|
x1 = focus.x - x_half
|
||||||
d = ImageDraw.Draw(im)
|
if x1 < 0:
|
||||||
rect = list(crop)
|
x1 = 0
|
||||||
rect[2] -= 1
|
elif x1 + settings.crop_width > im.width:
|
||||||
rect[3] -= 1
|
x1 = im.width - settings.crop_width
|
||||||
d.rectangle(rect, outline=GREEN)
|
|
||||||
if settings.destop_view_image:
|
y1 = focus.y - y_half
|
||||||
im.show()
|
if y1 < 0:
|
||||||
|
y1 = 0
|
||||||
return im.crop(tuple(crop))
|
elif y1 + settings.crop_height > im.height:
|
||||||
|
y1 = im.height - settings.crop_height
|
||||||
def focal_point(im, settings):
|
|
||||||
corner_points = image_corner_points(im, settings)
|
x2 = x1 + settings.crop_width
|
||||||
entropy_points = image_entropy_points(im, settings)
|
y2 = y1 + settings.crop_height
|
||||||
face_points = image_face_points(im, settings)
|
|
||||||
|
crop = [x1, y1, x2, y2]
|
||||||
total_points = len(corner_points) + len(entropy_points) + len(face_points)
|
|
||||||
|
if settings.annotate_image:
|
||||||
corner_weight = settings.corner_points_weight
|
d = ImageDraw.Draw(im)
|
||||||
entropy_weight = settings.entropy_points_weight
|
rect = list(crop)
|
||||||
face_weight = settings.face_points_weight
|
rect[2] -= 1
|
||||||
|
rect[3] -= 1
|
||||||
weight_pref_total = corner_weight + entropy_weight + face_weight
|
d.rectangle(rect, outline=GREEN)
|
||||||
|
if settings.destop_view_image:
|
||||||
# weight things
|
im.show()
|
||||||
pois = []
|
|
||||||
if weight_pref_total == 0 or total_points == 0:
|
return im.crop(tuple(crop))
|
||||||
return pois
|
|
||||||
|
def focal_point(im, settings):
|
||||||
pois.extend(
|
corner_points = image_corner_points(im, settings)
|
||||||
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (corner_weight/weight_pref_total) / (len(corner_points)/total_points) )) for p in corner_points ]
|
entropy_points = image_entropy_points(im, settings)
|
||||||
)
|
face_points = image_face_points(im, settings)
|
||||||
pois.extend(
|
|
||||||
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (entropy_weight/weight_pref_total) / (len(entropy_points)/total_points) )) for p in entropy_points ]
|
total_points = len(corner_points) + len(entropy_points) + len(face_points)
|
||||||
)
|
|
||||||
pois.extend(
|
corner_weight = settings.corner_points_weight
|
||||||
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (face_weight/weight_pref_total) / (len(face_points)/total_points) )) for p in face_points ]
|
entropy_weight = settings.entropy_points_weight
|
||||||
)
|
face_weight = settings.face_points_weight
|
||||||
|
|
||||||
average_point = poi_average(pois, settings)
|
weight_pref_total = corner_weight + entropy_weight + face_weight
|
||||||
|
|
||||||
if settings.annotate_image:
|
# weight things
|
||||||
d = ImageDraw.Draw(im)
|
pois = []
|
||||||
for f in face_points:
|
if weight_pref_total == 0 or total_points == 0:
|
||||||
d.rectangle(f.bounding(f.size), outline=RED)
|
return pois
|
||||||
for f in entropy_points:
|
|
||||||
d.rectangle(f.bounding(30), outline=BLUE)
|
pois.extend(
|
||||||
for poi in pois:
|
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (corner_weight/weight_pref_total) / (len(corner_points)/total_points) )) for p in corner_points ]
|
||||||
w = max(4, 4 * 0.5 * sqrt(poi.weight))
|
)
|
||||||
d.ellipse(poi.bounding(w), fill=BLUE)
|
pois.extend(
|
||||||
d.ellipse(average_point.bounding(25), outline=GREEN)
|
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (entropy_weight/weight_pref_total) / (len(entropy_points)/total_points) )) for p in entropy_points ]
|
||||||
|
)
|
||||||
return average_point
|
pois.extend(
|
||||||
|
[ PointOfInterest( p.x, p.y, weight=p.weight * ( (face_weight/weight_pref_total) / (len(face_points)/total_points) )) for p in face_points ]
|
||||||
|
)
|
||||||
def image_face_points(im, settings):
|
|
||||||
np_im = np.array(im)
|
average_point = poi_average(pois, settings)
|
||||||
gray = cv2.cvtColor(np_im, cv2.COLOR_BGR2GRAY)
|
|
||||||
|
if settings.annotate_image:
|
||||||
tries = [
|
d = ImageDraw.Draw(im)
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_eye.xml', 0.01 ],
|
for f in face_points:
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_frontalface_default.xml', 0.05 ],
|
d.rectangle(f.bounding(f.size), outline=RED)
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_profileface.xml', 0.05 ],
|
for f in entropy_points:
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt.xml', 0.05 ],
|
d.rectangle(f.bounding(30), outline=BLUE)
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt2.xml', 0.05 ],
|
for poi in pois:
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt_tree.xml', 0.05 ],
|
w = max(4, 4 * 0.5 * sqrt(poi.weight))
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_eye_tree_eyeglasses.xml', 0.05 ],
|
d.ellipse(poi.bounding(w), fill=BLUE)
|
||||||
[ f'{cv2.data.haarcascades}haarcascade_upperbody.xml', 0.05 ]
|
d.ellipse(average_point.bounding(25), outline=GREEN)
|
||||||
]
|
|
||||||
|
return average_point
|
||||||
for t in tries:
|
|
||||||
# print(t[0])
|
|
||||||
classifier = cv2.CascadeClassifier(t[0])
|
def image_face_points(im, settings):
|
||||||
minsize = int(min(im.width, im.height) * t[1]) # at least N percent of the smallest side
|
np_im = np.array(im)
|
||||||
try:
|
gray = cv2.cvtColor(np_im, cv2.COLOR_BGR2GRAY)
|
||||||
faces = classifier.detectMultiScale(gray, scaleFactor=1.1,
|
|
||||||
minNeighbors=7, minSize=(minsize, minsize), flags=cv2.CASCADE_SCALE_IMAGE)
|
tries = [
|
||||||
except:
|
[ f'{cv2.data.haarcascades}haarcascade_eye.xml', 0.01 ],
|
||||||
continue
|
[ f'{cv2.data.haarcascades}haarcascade_frontalface_default.xml', 0.05 ],
|
||||||
|
[ f'{cv2.data.haarcascades}haarcascade_profileface.xml', 0.05 ],
|
||||||
if len(faces) > 0:
|
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt.xml', 0.05 ],
|
||||||
rects = [[f[0], f[1], f[0] + f[2], f[1] + f[3]] for f in faces]
|
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt2.xml', 0.05 ],
|
||||||
return [PointOfInterest((r[0] +r[2]) // 2, (r[1] + r[3]) // 2, size=abs(r[0]-r[2])) for r in rects]
|
[ f'{cv2.data.haarcascades}haarcascade_frontalface_alt_tree.xml', 0.05 ],
|
||||||
return []
|
[ f'{cv2.data.haarcascades}haarcascade_eye_tree_eyeglasses.xml', 0.05 ],
|
||||||
|
[ f'{cv2.data.haarcascades}haarcascade_upperbody.xml', 0.05 ]
|
||||||
|
]
|
||||||
def image_corner_points(im, settings):
|
|
||||||
grayscale = im.convert("L")
|
for t in tries:
|
||||||
|
# print(t[0])
|
||||||
# naive attempt at preventing focal points from collecting at watermarks near the bottom
|
classifier = cv2.CascadeClassifier(t[0])
|
||||||
gd = ImageDraw.Draw(grayscale)
|
minsize = int(min(im.width, im.height) * t[1]) # at least N percent of the smallest side
|
||||||
gd.rectangle([0, im.height*.9, im.width, im.height], fill="#999")
|
try:
|
||||||
|
faces = classifier.detectMultiScale(gray, scaleFactor=1.1,
|
||||||
np_im = np.array(grayscale)
|
minNeighbors=7, minSize=(minsize, minsize), flags=cv2.CASCADE_SCALE_IMAGE)
|
||||||
|
except:
|
||||||
points = cv2.goodFeaturesToTrack(
|
continue
|
||||||
np_im,
|
|
||||||
maxCorners=100,
|
if len(faces) > 0:
|
||||||
qualityLevel=0.04,
|
rects = [[f[0], f[1], f[0] + f[2], f[1] + f[3]] for f in faces]
|
||||||
minDistance=min(grayscale.width, grayscale.height)*0.07,
|
return [PointOfInterest((r[0] +r[2]) // 2, (r[1] + r[3]) // 2, size=abs(r[0]-r[2])) for r in rects]
|
||||||
useHarrisDetector=False,
|
return []
|
||||||
)
|
|
||||||
|
|
||||||
if points is None:
|
def image_corner_points(im, settings):
|
||||||
return []
|
grayscale = im.convert("L")
|
||||||
|
|
||||||
focal_points = []
|
# naive attempt at preventing focal points from collecting at watermarks near the bottom
|
||||||
for point in points:
|
gd = ImageDraw.Draw(grayscale)
|
||||||
x, y = point.ravel()
|
gd.rectangle([0, im.height*.9, im.width, im.height], fill="#999")
|
||||||
focal_points.append(PointOfInterest(x, y, size=4))
|
|
||||||
|
np_im = np.array(grayscale)
|
||||||
return focal_points
|
|
||||||
|
points = cv2.goodFeaturesToTrack(
|
||||||
|
np_im,
|
||||||
def image_entropy_points(im, settings):
|
maxCorners=100,
|
||||||
landscape = im.height < im.width
|
qualityLevel=0.04,
|
||||||
portrait = im.height > im.width
|
minDistance=min(grayscale.width, grayscale.height)*0.07,
|
||||||
if landscape:
|
useHarrisDetector=False,
|
||||||
move_idx = [0, 2]
|
)
|
||||||
move_max = im.size[0]
|
|
||||||
elif portrait:
|
if points is None:
|
||||||
move_idx = [1, 3]
|
return []
|
||||||
move_max = im.size[1]
|
|
||||||
else:
|
focal_points = []
|
||||||
return []
|
for point in points:
|
||||||
|
x, y = point.ravel()
|
||||||
e_max = 0
|
focal_points.append(PointOfInterest(x, y, size=4))
|
||||||
crop_current = [0, 0, settings.crop_width, settings.crop_height]
|
|
||||||
crop_best = crop_current
|
return focal_points
|
||||||
while crop_current[move_idx[1]] < move_max:
|
|
||||||
crop = im.crop(tuple(crop_current))
|
|
||||||
e = image_entropy(crop)
|
def image_entropy_points(im, settings):
|
||||||
|
landscape = im.height < im.width
|
||||||
if (e > e_max):
|
portrait = im.height > im.width
|
||||||
e_max = e
|
if landscape:
|
||||||
crop_best = list(crop_current)
|
move_idx = [0, 2]
|
||||||
|
move_max = im.size[0]
|
||||||
crop_current[move_idx[0]] += 4
|
elif portrait:
|
||||||
crop_current[move_idx[1]] += 4
|
move_idx = [1, 3]
|
||||||
|
move_max = im.size[1]
|
||||||
x_mid = int(crop_best[0] + settings.crop_width/2)
|
else:
|
||||||
y_mid = int(crop_best[1] + settings.crop_height/2)
|
return []
|
||||||
|
|
||||||
return [PointOfInterest(x_mid, y_mid, size=25)]
|
e_max = 0
|
||||||
|
crop_current = [0, 0, settings.crop_width, settings.crop_height]
|
||||||
|
crop_best = crop_current
|
||||||
def image_entropy(im):
|
while crop_current[move_idx[1]] < move_max:
|
||||||
# greyscale image entropy
|
crop = im.crop(tuple(crop_current))
|
||||||
# band = np.asarray(im.convert("L"))
|
e = image_entropy(crop)
|
||||||
band = np.asarray(im.convert("1"), dtype=np.uint8)
|
|
||||||
hist, _ = np.histogram(band, bins=range(0, 256))
|
if (e > e_max):
|
||||||
hist = hist[hist > 0]
|
e_max = e
|
||||||
return -np.log2(hist / hist.sum()).sum()
|
crop_best = list(crop_current)
|
||||||
|
|
||||||
|
crop_current[move_idx[0]] += 4
|
||||||
def poi_average(pois, settings):
|
crop_current[move_idx[1]] += 4
|
||||||
weight = 0.0
|
|
||||||
x = 0.0
|
x_mid = int(crop_best[0] + settings.crop_width/2)
|
||||||
y = 0.0
|
y_mid = int(crop_best[1] + settings.crop_height/2)
|
||||||
for poi in pois:
|
|
||||||
weight += poi.weight
|
return [PointOfInterest(x_mid, y_mid, size=25)]
|
||||||
x += poi.x * poi.weight
|
|
||||||
y += poi.y * poi.weight
|
|
||||||
avg_x = round(x / weight)
|
def image_entropy(im):
|
||||||
avg_y = round(y / weight)
|
# greyscale image entropy
|
||||||
|
# band = np.asarray(im.convert("L"))
|
||||||
return PointOfInterest(avg_x, avg_y)
|
band = np.asarray(im.convert("1"), dtype=np.uint8)
|
||||||
|
hist, _ = np.histogram(band, bins=range(0, 256))
|
||||||
|
hist = hist[hist > 0]
|
||||||
class PointOfInterest:
|
return -np.log2(hist / hist.sum()).sum()
|
||||||
def __init__(self, x, y, weight=1.0, size=10):
|
|
||||||
self.x = x
|
|
||||||
self.y = y
|
def poi_average(pois, settings):
|
||||||
self.weight = weight
|
weight = 0.0
|
||||||
self.size = size
|
x = 0.0
|
||||||
|
y = 0.0
|
||||||
def bounding(self, size):
|
for poi in pois:
|
||||||
return [
|
weight += poi.weight
|
||||||
self.x - size//2,
|
x += poi.x * poi.weight
|
||||||
self.y - size//2,
|
y += poi.y * poi.weight
|
||||||
self.x + size//2,
|
avg_x = round(x / weight)
|
||||||
self.y + size//2
|
avg_y = round(y / weight)
|
||||||
]
|
|
||||||
|
return PointOfInterest(avg_x, avg_y)
|
||||||
|
|
||||||
class Settings:
|
|
||||||
def __init__(self, crop_width=512, crop_height=512, corner_points_weight=0.5, entropy_points_weight=0.5, face_points_weight=0.5, annotate_image=False):
|
def is_landscape(w, h):
|
||||||
self.crop_width = crop_width
|
return w > h
|
||||||
self.crop_height = crop_height
|
|
||||||
self.corner_points_weight = corner_points_weight
|
|
||||||
self.entropy_points_weight = entropy_points_weight
|
def is_portrait(w, h):
|
||||||
self.face_points_weight = entropy_points_weight
|
return h > w
|
||||||
self.annotate_image = annotate_image
|
|
||||||
|
|
||||||
|
def is_square(w, h):
|
||||||
|
return w == h
|
||||||
|
|
||||||
|
|
||||||
|
class PointOfInterest:
|
||||||
|
def __init__(self, x, y, weight=1.0, size=10):
|
||||||
|
self.x = x
|
||||||
|
self.y = y
|
||||||
|
self.weight = weight
|
||||||
|
self.size = size
|
||||||
|
|
||||||
|
def bounding(self, size):
|
||||||
|
return [
|
||||||
|
self.x - size//2,
|
||||||
|
self.y - size//2,
|
||||||
|
self.x + size//2,
|
||||||
|
self.y + size//2
|
||||||
|
]
|
||||||
|
|
||||||
|
|
||||||
|
class Settings:
|
||||||
|
def __init__(self, crop_width=512, crop_height=512, corner_points_weight=0.5, entropy_points_weight=0.5, face_points_weight=0.5, annotate_image=False):
|
||||||
|
self.crop_width = crop_width
|
||||||
|
self.crop_height = crop_height
|
||||||
|
self.corner_points_weight = corner_points_weight
|
||||||
|
self.entropy_points_weight = entropy_points_weight
|
||||||
|
self.face_points_weight = entropy_points_weight
|
||||||
|
self.annotate_image = annotate_image
|
||||||
self.destop_view_image = False
|
self.destop_view_image = False
|
||||||
Loading…
Reference in New Issue