inkstitch/lib/stitches/running_stitch.py

397 wiersze
15 KiB
Python

# Authors: see git history
#
# Copyright (c) 2010 Authors
# Licensed under the GNU GPL version 3.0 or later. See the file LICENSE for details.
import math
import typing
from copy import copy
from math import tau
import numpy as np
from shapely import geometry as shgeo
from ..utils import prng
from ..utils.geometry import Point
from ..utils.threading import check_stop_flag
""" Utility functions to produce running stitches. """
def lerp(a, b, t: float) -> float:
return (1 - t) * a + t * b
def split_segment_even_n(a, b, segments: int, jitter_sigma: float = 0.0, random_seed=None) -> typing.List[shgeo.Point]:
if segments <= 1:
return []
line = shgeo.LineString((a, b))
splits = np.array(range(1, segments)) / segments
if random_seed is not None:
jitters = (prng.n_uniform_floats(len(splits), random_seed) * 2) - 1
splits = splits + jitters * (jitter_sigma / segments)
# sort the splits in case a bad roll transposes any of them
return [line.interpolate(x, normalized=True) for x in sorted(splits)]
def split_segment_even_dist(a, b, max_length: float, jitter_sigma: float = 0.0, random_seed=None) -> typing.List[shgeo.Point]:
distance = shgeo.Point(a).distance(shgeo.Point(b))
segments = math.ceil(distance / max_length)
return split_segment_even_n(a, b, segments, jitter_sigma, random_seed)
def split_segment_random_phase(a, b, length: float, length_sigma: float, random_seed: str) -> typing.List[shgeo.Point]:
line = shgeo.LineString([a, b])
progress = length * prng.uniform_floats(random_seed, "phase")[0]
splits = [progress]
distance = line.length
if progress >= distance:
return []
for x in prng.iter_uniform_floats(random_seed):
progress += length * (1 + length_sigma * (x - 0.5) * 2)
if progress >= distance:
break
splits.append(progress)
return [line.interpolate(x, normalized=False) for x in splits]
def split_segment_stagger_phase(a, b, segment_length: float, num_staggers: int, this_segment_num: int, min=0, max=None) \
-> typing.List[shgeo.Point]:
line = shgeo.LineString([a, b])
distance = line.length
stagger_phase = (this_segment_num / num_staggers) % 1
stagger_offset = stagger_phase * segment_length
if max is None:
max = distance
splits = []
progress = stagger_offset
while progress < distance:
if progress > min and progress < max:
splits.append(progress)
progress += segment_length
return [line.interpolate(x, normalized=False) for x in splits]
class AngleInterval():
# Modular interval containing either the entire circle or less than half of it
# partially based on https://fgiesen.wordpress.com/2015/09/24/intervals-in-modular-arithmetic/
def __init__(self, a: float, b: float, all: bool = False):
self.all = all
self.a = a
self.b = b
@staticmethod
def all():
return AngleInterval(0, math.tau, True)
@staticmethod
def fromBall(p: Point, epsilon: float):
d = p.length()
if d <= epsilon:
return AngleInterval.all()
center = p.angle()
delta = math.asin(epsilon / d)
return AngleInterval(center - delta, center + delta)
@staticmethod
def fromSegment(a: Point, b: Point):
angleA = a.angle()
angleB = b.angle()
diff = (angleB - angleA) % tau
if diff == 0 or diff == math.pi:
return None
elif diff < math.pi:
return AngleInterval(angleA - 1e-6, angleB + 1e-6)
# slightly larger than normal to avoid rounding error when this method is used in cutSegment
else:
return AngleInterval(angleB - 1e-6, angleA + 1e-6)
def containsAngle(self, angle: float):
if self.all:
return True
return (angle - self.a) % tau <= (self.b - self.a) % tau
def containsPoint(self, p: Point):
return self.containsAngle(math.atan2(p.y, p.x))
def intersect(self, other):
# assume that each interval contains less than half the circle (or all of it)
if other is None:
return None
elif self.all:
return other
elif other.all:
return self
elif self.containsAngle(other.a):
if other.containsAngle(self.b):
return AngleInterval(other.a, self.b)
else:
return AngleInterval(other.a, other.b)
elif other.containsAngle(self.a):
if self.containsAngle(other.b):
return AngleInterval(self.a, other.b)
else:
return AngleInterval(self.a, self.b)
else:
return None
def cutSegment(self, origin: Point, a: Point, b: Point):
if self.all:
return None
segArc = AngleInterval.fromSegment(a - origin, b - origin)
if segArc is None:
return a # b is exactly behind origin from a
if segArc.containsAngle(self.a):
return cut_segment_with_angle(origin, self.a, a, b)
elif segArc.containsAngle(self.b):
return cut_segment_with_angle(origin, self.b, a, b)
else:
return None
def cut_segment_with_angle(origin: Point, angle: float, a: Point, b: Point) -> Point:
# Assumes the crossing is inside the segment
p = a - origin
d = b - a
c = Point(math.cos(angle), math.sin(angle))
t = (p.y*c.x - p.x*c.y) / (d.x*c.y - d.y*c.x)
if t < -0.000001 or t > 1.000001:
raise Exception("cut_segment_with_angle returned a parameter of {0} with points {1} {2} and cut line {3} ".format(t, p, b-origin, c))
return a + d*t
def cut_segment_with_circle(origin: Point, r: float, a: Point, b: Point) -> Point:
# assumes that a is inside the circle and b is outside
p = a - origin
d = b - a
# inner products
p2 = p * p
d2 = d * d
r2 = r * r
pd = p * d
# r2 = p2 + 2*pd*t + d2*t*t, quadratic formula
t = (math.sqrt(pd*pd + r2*d2 - p2*d2) - pd) / d2
if t < -0.000001 or t > 1.000001:
raise Exception("cut_segment_with_circle returned a parameter of {0}".format(t))
return a + d*t
def take_stitch(start: Point, points: typing.Sequence[Point], idx: int, stitch_length: float, tolerance: float):
# Based on a single step of the Zhao-Saalfeld curve simplification algorithm.
# https://cartogis.org/docs/proceedings/archive/auto-carto-13/pdf/linear-time-sleeve-fitting-polyline-simplification-algorithms.pdf
# Adds early termination condition based on stitch length.
if idx >= len(points):
return None, None
sleeve = AngleInterval.all()
last = start
for i in range(idx, len(points)):
p = points[i]
if sleeve.containsPoint(p - start):
if start.distance(p) < stitch_length:
sleeve = sleeve.intersect(AngleInterval.fromBall(p - start, tolerance))
last = p
continue
else:
cut = cut_segment_with_circle(start, stitch_length, last, p)
return cut, i
else:
cut = sleeve.cutSegment(start, last, p)
if start.distance(cut) > stitch_length:
cut = cut_segment_with_circle(start, stitch_length, last, p)
return cut, i
return points[-1], None
def stitch_curve_evenly(points: typing.Sequence[Point], stitch_length: float, tolerance: float):
# Will split a straight line into even-length stitches while still handling curves correctly.
# Includes end point but not start point.
if len(points) < 2:
return []
distLeft = [0] * len(points)
for i in reversed(range(0, len(points) - 1)):
distLeft[i] = distLeft[i + 1] + points[i].distance(points[i+1])
i = 1
last = points[0]
stitches = []
while i is not None and i < len(points):
d = last.distance(points[i]) + distLeft[i]
if d == 0:
return stitches
stitch_len = d / math.ceil(d / stitch_length) + 0.000001 # correction for rounding error
stitch, newidx = take_stitch(last, points, i, stitch_len, tolerance)
i = newidx
if stitch is not None:
stitches.append(stitch)
last = stitch
return stitches
def stitch_curve_randomly(points: typing.Sequence[Point], stitch_length: float, tolerance: float, stitch_length_sigma: float, random_seed: str):
min_stitch_length = max(0, stitch_length * (1 - stitch_length_sigma))
max_stitch_length = stitch_length * (1 + stitch_length_sigma)
# Will split a straight line into stitches of random length within the range.
# Attempts to randomize phase so that the distribution of outputs does not depend on direction.
# Includes end point but not start point.
if len(points) < 2:
return []
i = 1
last = points[0]
last_shortened = 0.0
stitches = []
rand_iter = iter(prng.iter_uniform_floats(random_seed))
while i is not None and i < len(points):
r = next(rand_iter)
# If the last stitch was shortened due to tolerance (or this is the first stitch),
# reduce the lower length limit to randomize the phase. This prevents moiré and asymmetry.
stitch_len = lerp(last_shortened, 1.0, r) * lerp(min_stitch_length, max_stitch_length, r)
stitch, newidx = take_stitch(last, points, i, stitch_len, tolerance)
i = newidx
if stitch is not None:
stitches.append(stitch)
last_shortened = min(last.distance(stitch) / stitch_len, 1.0)
last = stitch
return stitches
def path_to_curves(points: typing.List[Point], min_len: float):
# split a path at obvious corner points so that they get stitched exactly
# min_len controls the minimum length after splitting for which it won't split again,
# which is used to avoid creating large numbers of corner points when encouintering micro-messes.
if len(points) < 3:
return [points]
curves = []
last = 0
last_seg = points[1] - points[0]
seg_len = last_seg.length()
for i in range(1, len(points) - 1):
# vectors of the last and next segments
a = last_seg
b = points[i + 1] - points[i]
aabb = (a * a) * (b * b)
abab = (a * b) * abs(a * b)
# Test if the turn angle from vectors a to b is more than 45 degrees.
# Optimized version of checking if cos(angle(a,b)) <= sqrt(0.5) and is defined
if aabb > 0 and abab <= 0.5 * aabb:
if seg_len >= min_len:
curves.append(points[last: i + 1])
last = i
seg_len = 0
if b * b > 0:
last_seg = b
seg_len += b.length()
curves.append(points[last:])
return curves
def even_running_stitch(points, stitch_length, tolerance):
# Turn a continuous path into a running stitch with as close to even stitch length as possible
# (including the first and last segments), keeping it within the tolerance of the path.
# This should not be used for stitching tightly-spaced parallel curves
# as it tends to produce ugly moiré effects in those situations.
# In these situations, random_running_stitch sould be used even if the maximum stitch length range is a single value.
if not points:
return
stitches = [points[0]]
for curve in path_to_curves(points, 2 * tolerance):
# segments longer than twice the tolerance will usually be forced by it, so set that as the minimum for corner detection
check_stop_flag()
stitches.extend(stitch_curve_evenly(curve, stitch_length, tolerance))
return stitches
def random_running_stitch(points, stitch_length, tolerance, stitch_length_sigma, random_seed):
# Turn a continuous path into a running stitch with randomized phase and stitch length,
# keeping it within the tolerance of the path.
# This is suitable for tightly-spaced parallel curves.
if not points:
return
stitches = [points[0]]
for i, curve in enumerate(path_to_curves(points, 2 * tolerance)):
# segments longer than twice the tolerance will usually be forced by it, so set that as the minimum for corner detection
check_stop_flag()
stitches.extend(stitch_curve_randomly(curve, stitch_length, tolerance, stitch_length_sigma, prng.join_args(random_seed, i)))
return stitches
def running_stitch(points, stitch_length, tolerance, is_random, stitch_length_sigma, random_seed):
# running stitch with a choice of algorithm
if is_random:
return random_running_stitch(points, stitch_length, tolerance, stitch_length_sigma, random_seed)
else:
return even_running_stitch(points, stitch_length, tolerance)
def bean_stitch(stitches, repeats, tags_to_ignore=None):
"""Generate bean stitch from a set of stitches.
"Bean" stitch is made by backtracking each stitch to make it heavier. A
simple bean stitch would be two stitches forward, one stitch back, two
stitches forward, etc. This would result in each stitch being tripled.
We'll say that the above counts as 1 repeat. Backtracking each stitch
repeatedly will result in a heavier bean stitch. There will always be
an odd number of threads piled up for each stitch.
Repeats is a list of a repeated pattern e.g. [0, 1, 3] doesn't repeat the first stitch,
goes back and forth on the second stitch, goes goes 3 times back and forth on the third stitch,
and starts the pattern again by not repeating the fourth stitch, etc.
"""
if len(stitches) < 2:
return stitches
repeat_list_length = len(repeats)
new_stitches = [stitches[0]]
for i, stitch in enumerate(stitches[1:]):
repeat_list_pos = i % repeat_list_length
new_stitches.append(stitch)
# ignore stitches with specified tags
if tags_to_ignore and set(tags_to_ignore).intersection(stitch.tags):
continue
for i in range(repeats[repeat_list_pos]):
new_stitches.extend(copy(new_stitches[-2:]))
return new_stitches
def zigzag_stitch(stitches, zigzag_spacing, stroke_width, pull_compensation):
# Move the points left and right. Consider each pair
# of points in turn, and move perpendicular to them,
# alternating left and right.
stroke_width = stroke_width + pull_compensation
offset = stroke_width / 2.0
for i in range(len(stitches) - 1):
start = stitches[i]
end = stitches[i + 1]
# sometimes the stitch results into zero length which cause a division by zero error
# ignoring this leads to a slightly bad result, but that is better than no output
if (end - start).length() == 0:
continue
segment_direction = (end - start).unit()
zigzag_direction = segment_direction.rotate_left()
if i % 2 == 1:
zigzag_direction *= -1
stitches[i] += zigzag_direction * offset
return stitches