ii-pix/dither_apply.pyx

105 lines
4.0 KiB
Cython
Raw Normal View History

2021-01-10 22:12:14 +00:00
# cython: infer_types=True
cimport cython
import numpy as np
# from cython.parallel import prange
from cython.view cimport array as cvarray
2021-01-11 18:55:37 +00:00
from libc.stdlib cimport malloc, free
2021-01-10 22:12:14 +00:00
cdef float clip(float a, float min_value, float max_value) nogil:
return min(max(a, min_value), max_value)
#@cython.boundscheck(False)
#@cython.wraparound(False)
def apply_one_line(float[:, :, ::1] pattern, int el, int er, int xl, int xr, int y, float[:, ::1] image,
2021-01-10 22:12:14 +00:00
float[::1] quant_error):
cdef int i, j
2021-01-11 18:55:37 +00:00
cdef float *error = <float *> malloc(pattern.shape[1] * quant_error.shape[0] * sizeof(float))
#cdef float[:, ::1] error = cvarray(
# shape=(pattern.shape[1], quant_error.shape[0]), itemsize=sizeof(float), format="f")
2021-01-10 22:12:14 +00:00
for i in range(pattern.shape[1]):
for j in range(quant_error.shape[0]):
2021-01-11 18:55:37 +00:00
error[i * quant_error.shape[0] + j] = pattern[0, i, 0] * quant_error[j]
2021-01-10 22:12:14 +00:00
for i in range(xr - xl):
for j in range(3):
image[xl+i, j] = clip(image[xl + i, j] + error[(el + i) * quant_error.shape[0] + j], 0, 255)
2021-01-11 18:55:37 +00:00
free(error)
2021-01-10 22:12:14 +00:00
# XXX cythonize
def apply(pattern, int el, int er, int xl, int xr, int et, int eb, int yt, int yb, image, quant_error):
error = pattern * quant_error.reshape((1, 1, 3))
# We could avoid clipping here, i.e. allow RGB values to extend beyond
# 0..255 to capture a larger range of residual error. This is faster
# but seems to reduce image quality.
# XXX extend image region to avoid need for boundary box clipping
image[yt:yb, xl:xr, :] = np.clip(
image[yt:yb, xl:xr, :] + error[et:eb, el:er, :], 0, 255)
def x_dither_bounds(dither, screen, int x):
cdef int el = max(dither.ORIGIN[1] - x, 0)
cdef int er = min(dither.PATTERN.shape[1], screen.X_RES - 1 - x)
cdef int xl = x - dither.ORIGIN[1] + el
cdef int xr = x - dither.ORIGIN[1] + er
return el, er, xl, xr
def y_dither_bounds(dither, screen, int y):
cdef int et = max(dither.ORIGIN[0] - y, 0)
cdef int eb = min(dither.PATTERN.shape[0], screen.Y_RES - 1 - y)
cdef int yt = y - dither.ORIGIN[0] + et
cdef int yb = y - dither.ORIGIN[0] + eb
return et, eb, yt, yb
def dither_lookahead(
screen, float[:,:,::1] image_rgb, dither, differ, int x, int y, char[:, ::1] options_4bit,
float[:, :, ::1] options_rgb, int lookahead):
el, er, xl, xr = x_dither_bounds(dither, screen, x)
# X coord value of larger of dither bounding box or lookahead horizon
xxr = min(max(x + lookahead, xr), screen.X_RES)
# copies of input pixels so we can dither in bulk
# Leave enough space so we can dither the last of our lookahead pixels
lah_image_rgb = np.zeros(
(2 ** lookahead, lookahead + xr - xl, 3), dtype=np.float32)
lah_image_rgb[:, 0:xxr - x, :] = np.copy(image_rgb[y, x:xxr, :])
#cdef float[:, :, ::1] lah_image_rgb_view = lah_image_rgb
#cdef float[:, :, ::1] options_rgb_view = options_rgb
cdef float[:, ::] output_pixels
cdef float[:, ::1] quant_error
cdef int i, j
for i in range(xxr - x):
# options_rgb choices are fixed, but we can still distribute
# quantization error from having made these choices, in order to compute
# the total error
input_pixels = np.copy(lah_image_rgb[:, i, :])
output_pixels = options_rgb[:, i, :]
quant_error = input_pixels - output_pixels
# Don't update the input at position x (since we've already chosen
# fixed outputs), but do propagate quantization errors to positions >x
# so we can compensate for how good/bad these choices were
el, er, xl, xr = x_dither_bounds(dither, screen, i)
for j in range(2 ** lookahead):
apply_one_line(dither.PATTERN, el, er, xl, xr, 0, lah_image_rgb[j, :, :], quant_error[j])
error = differ.distance(np.clip(
lah_image_rgb[:, 0:lookahead, :], 0, 255), options_4bit)
total_error = np.sum(np.power(error, 2), axis=1)
best = np.argmin(total_error)
return options_4bit[best, 0], options_rgb[best, 0, :]