ii-pix/dither.pyx
2021-03-15 17:22:14 +00:00

340 lines
13 KiB
Cython

# cython: infer_types=True
# cython: profile=True
cimport cython
import functools
import numpy as np
from cython.view cimport array as cvarray
from libc.stdlib cimport malloc, free
# TODO: use a cdef class
# C representation of dither_pattern.DitherPattern data, for efficient access.
cdef struct Dither:
float* pattern # Flattened dither pattern
int x_shape
int y_shape
int x_origin
int y_origin
cdef float clip(float a, float min_value, float max_value) nogil:
return min(max(a, min_value), max_value)
# Compute left-hand bounding box for dithering at horizontal position x.
cdef int dither_bounds_xl(Dither *dither, int x) nogil:
cdef int el = max(dither.x_origin - x, 0)
cdef int xl = x - dither.x_origin + el
return xl
#Compute right-hand bounding box for dithering at horizontal position x.
cdef int dither_bounds_xr(Dither *dither, int x_res, int x) nogil:
cdef int er = min(dither.x_shape, x_res - x)
cdef int xr = x - dither.x_origin + er
return xr
# Compute upper bounding box for dithering at vertical position y.
cdef int dither_bounds_yt(Dither *dither, int y) nogil:
cdef int et = max(dither.y_origin - y, 0)
cdef int yt = y - dither.y_origin + et
return yt
# Compute lower bounding box for dithering at vertical position y.
cdef int dither_bounds_yb(Dither *dither, int y_res, int y) nogil:
cdef int eb = min(dither.y_shape, y_res - y)
cdef int yb = y - dither.y_origin + eb
return yb
@cython.boundscheck(False)
@cython.wraparound(False)
@functools.lru_cache(None)
def lookahead_options(object screen, int lookahead, unsigned char last_pixel_nbit, int x):
"""Compute all possible n-bit palette values for upcoming pixels, given x coord and state of n pixels to the left.
Args:
screen: python screen.Screen object
lookahead: how many pixels to lookahead
last_pixel_nbit: n-bit value representing n pixels to left of current position, which determine available
colours.
x: current x position
Returns: matrix of size (2**lookahead, lookahead) containing all 2**lookahead possible vectors of n-bit palette
values accessible at positions x .. x + lookahead
"""
cdef unsigned char[:, ::1] options_nbit = np.empty((2 ** lookahead, lookahead), dtype=np.uint8)
cdef int i, j, xx, p
cdef unsigned char output_pixel_nbit
cdef unsigned char[::1] palette_choices_nbit
cdef object palette = screen.palette
cdef dict palette_rgb = palette.RGB
for i in range(2 ** lookahead):
output_pixel_nbit = last_pixel_nbit
for j in range(lookahead):
xx = x + j
# Two possible n-bit palette choices at position xx, given state of n pixels to left.
# TODO: port screen.py to pyx
palette_choices_nbit = screen.pixel_palette_options(output_pixel_nbit, xx)
output_pixel_nbit = palette_choices_nbit[(i & (1 << j)) >> j]
options_nbit[i, j] = output_pixel_nbit
return options_nbit
# Look ahead a number of pixels and compute choice for next pixel with lowest total squared error after dithering.
#
# Args:
# dither: error diffusion pattern to apply
# palette_rgb: matrix of all n-bit colour palette RGB values
# image_rgb: RGB image in the process of dithering
# x: current horizontal screen position
# y: current vertical screen position
# options_nbit: matrix of (2**lookahead, lookahead) possible n-bit colour choices at positions x .. x + lookahead
# lookahead: how many horizontal pixels to look ahead
# distances: matrix of (24-bit RGB, n-bit palette) perceptual colour distances
# x_res: horizontal screen resolution
#
# Returns: index from 0 .. 2**lookahead into options_nbit representing best available choice for position (x,y)
#
@cython.boundscheck(False)
@cython.wraparound(False)
cdef int dither_lookahead(Dither* dither, float[:, ::1] palette_rgb,
float[:, :, ::1] image_rgb, int x, int y, unsigned char[:, ::1] options_nbit, int lookahead,
unsigned char[:, ::1] distances, int x_res):
cdef int i, j, k, l
cdef float[3] quant_error
cdef unsigned char bit4
cdef int best
cdef int best_error = 2**31-1
cdef int total_error
cdef long flat, dist
cdef long r, g, b
# Don't bother dithering past the lookahead horizon or edge of screen.
cdef int xxr = min(x + lookahead, x_res)
cdef int lah_shape1 = xxr - x
cdef int lah_shape2 = 3
cdef float *lah_image_rgb = <float *> malloc(lah_shape1 * lah_shape2 * sizeof(float))
# For each 2**lookahead possibilities for the on/off state of the next lookahead pixels, apply error diffusion
# and compute the total squared error to the source image. Since we only have two possible colours for each
# given pixel (dependent on the state already chosen for pixels to the left), we need to look beyond local minima.
# i.e. it might be better to make a sub-optimal choice for this pixel if it allows access to much better pixel
# colours at later positions.
for i in range(1 << lookahead):
# Working copy of input pixels
for j in range(xxr - x):
for k in range(3):
lah_image_rgb[j * lah_shape2 + k] = image_rgb[y, x+j, k]
total_error = 0
for j in range(xxr - x):
xl = dither_bounds_xl(dither, j)
xr = dither_bounds_xr(dither, xxr - x, j)
# We don't update the input at position x (since we've already chosen
# fixed outputs), but we do propagate quantization errors to positions >x
# so we can compensate for how good/bad these choices were. i.e. the
# options_rgb choices are fixed, but we can still distribute quantization error
# from having made these choices, in order to compute the total error.
for k in range(3):
quant_error[k] = lah_image_rgb[j * lah_shape2 + k] - palette_rgb[options_nbit[i,j], k]
apply_one_line(dither, xl, xr, j, lah_image_rgb, lah_shape2, quant_error)
r = <long>lah_image_rgb[j * lah_shape2 + 0]
g = <long>lah_image_rgb[j * lah_shape2 + 1]
b = <long>lah_image_rgb[j * lah_shape2 + 2]
flat = (r << 16) + (g << 8) + b
bit4 = options_nbit[i, j]
dist = distances[flat, bit4]
total_error += dist * dist
if total_error >= best_error:
break
if total_error < best_error:
best_error = total_error
best = i
free(lah_image_rgb)
return best
# Perform error diffusion to a single image row.
#
# Args:
# dither: dither pattern to apply
# xl: lower x bounding box
# xr: upper x bounding box
# x: starting horizontal position to apply error diffusion
# image: array of shape (image_shape1, 3) representing RGB pixel data for a single image line, to be mutated.
# image_shape1: horizontal dimension of image
# quant_error: RGB quantization error to be diffused
#
cdef void apply_one_line(Dither* dither, int xl, int xr, int x, float[] image, int image_shape1,
float[] quant_error) nogil:
cdef int i, j
cdef float error_fraction
for i in range(xl, xr):
error_fraction = dither.pattern[i - x + dither.x_origin]
for j in range(3):
image[i * image_shape1 + j] = clip(image[i * image_shape1 + j] + error_fraction * quant_error[j], 0, 255)
# Perform error diffusion across multiple image rows.
#
# Args:
# dither: dither pattern to apply
# x_res: horizontal image resolution
# y_res: vertical image resolution
# x: starting horizontal position to apply error diffusion
# y: starting vertical position to apply error diffusion
# image: RGB pixel data, to be mutated
# quant_error: RGB quantization error to be diffused
#
@cython.boundscheck(False)
@cython.wraparound(False)
cdef void apply(Dither* dither, int x_res, int y_res, int x, int y, float[:,:,::1] image, float[] quant_error):
cdef int i, j, k
cdef int yt = dither_bounds_yt(dither, y)
cdef int yb = dither_bounds_yb(dither, y_res, y)
cdef int xl = dither_bounds_xl(dither, x)
cdef int xr = dither_bounds_xr(dither, x_res, x)
cdef float error_fraction
# 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.
# TODO: is this still true?
for i in range(yt, yb):
for j in range(xl, xr):
error_fraction = dither.pattern[(i - y) * dither.x_shape + j - x + dither.x_origin]
for k in range(3):
image[i,j,k] = clip(image[i,j,k] + error_fraction * quant_error[k], 0, 255)
# Compute closest colour from array of candidate n-bit colour palette values.
#
# Args:
# pixel_rgb: source RGB colour value to be matched
# options_nbit: array of candidate n-bit colour palette values
# distances: matrix of (24-bit RGB value, n-bit colour value) perceptual colour differences
#
# Returns:
# index of options_nbit entry having lowest distance value
#
@cython.boundscheck(False)
@cython.wraparound(False)
cdef unsigned char find_nearest_colour(float[::1] pixel_rgb, unsigned char[::1] options_nbit,
unsigned char[:, ::1] distances):
cdef int best, dist
cdef unsigned char bit4
cdef int best_dist = 2**8
cdef long flat
for i in range(options_nbit.shape[0]):
flat = (<long>pixel_rgb[0] << 16) + (<long>pixel_rgb[1] << 8) + <long>pixel_rgb[2]
bit4 = options_nbit[i]
dist = distances[flat, bit4]
if dist < best_dist:
best_dist = dist
best = i
return options_nbit[best]
# Dither a source image
#
# Args:
# screen: screen.Screen object
# image_rgb: input RGB image
# dither: dither_pattern.DitherPattern to apply during dithering
# lookahead: how many x positions to look ahead to optimize colour choices
# verbose: whether to output progress during image conversion
#
# Returns: tuple of n-bit output image array and RGB output image array
#
@cython.boundscheck(False)
@cython.wraparound(False)
def dither_image(screen, float[:, :, ::1] image_rgb, dither, int lookahead, unsigned char verbose):
cdef int y, x, i
cdef float[3] input_pixel_rgb
cdef float[3] quant_error
cdef unsigned char [:, ::1] options_nbit
cdef float[:, :, ::1] options_rgb
cdef unsigned char [:, ::1] lookahead_palette_choices_nbit
cdef unsigned char [::1] palette_choices_nbit
cdef unsigned char output_pixel_nbit
cdef float[::1] output_pixel_rgb
# Hoist some python attribute accesses into C variables for efficient access during the main loop
cdef int yres = screen.Y_RES
cdef int xres = screen.X_RES
cdef float[:, ::1] palette_rgb = np.zeros((len(screen.palette.RGB), 3), dtype=np.float32)
for i in screen.palette.RGB.keys():
for j in range(3):
palette_rgb[i, j] = screen.palette.RGB[i][j]
cdef (unsigned char)[:, ::1] distances = screen.palette.distances
cdef Dither cdither
cdither.y_shape = dither.PATTERN.shape[0]
cdither.x_shape = dither.PATTERN.shape[1]
cdither.y_origin = dither.ORIGIN[0]
cdither.x_origin = dither.ORIGIN[1]
# TODO: should be just as efficient to use a memoryview?
cdither.pattern = <float *> malloc(cdither.x_shape * cdither.y_shape * sizeof(float))
for i in range(cdither.y_shape):
for j in range(cdither.x_shape):
cdither.pattern[i * cdither.x_shape + j] = dither.PATTERN[i, j]
cdef (unsigned char)[:, ::1] image_nbit = np.empty(
(image_rgb.shape[0], image_rgb.shape[1]), dtype=np.uint8)
for y in range(yres):
if verbose:
print("%d/%d" % (y, yres))
output_pixel_nbit = 0
for x in range(xres):
for i in range(3):
input_pixel_rgb[i] = image_rgb[y,x,i]
if lookahead:
# Compute all possible 2**N choices of n-bit pixel colours for positions x .. x + lookahead
lookahead_palette_choices_nbit = lookahead_options(screen, lookahead, output_pixel_nbit, x % 4)
# Apply error diffusion for each of these 2**N choices, and compute which produces the closest match
# to the source image over the succeeding N pixels
best_idx = dither_lookahead(
&cdither, palette_rgb, image_rgb, x, y, lookahead_palette_choices_nbit, lookahead, distances,
xres)
output_pixel_nbit = lookahead_palette_choices_nbit[best_idx, 0]
else:
# Choose the closest colour among the available n-bit palette options
palette_choices_nbit = screen.pixel_palette_options(output_pixel_nbit, x)
output_pixel_nbit = find_nearest_colour(input_pixel_rgb, palette_choices_nbit, distances)
# Apply error diffusion from chosen output pixel value
output_pixel_rgb = palette_rgb[output_pixel_nbit]
for i in range(3):
quant_error[i] = input_pixel_rgb[i] - output_pixel_rgb[i]
image_nbit[y, x] = output_pixel_nbit
apply(&cdither, xres, yres, x, y, image_rgb, quant_error)
for i in range(3):
image_rgb[y, x, i] = output_pixel_rgb[i]
free(cdither.pattern)
return image_nbit, np.array(image_rgb)