openmv/tools/gen_invariant.py
2018-05-25 20:33:50 +02:00

144 lines
5.7 KiB
Python

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# https://en.wikipedia.org/wiki/SRGB (The reverse transformation)
def lin(c):
return 100.0 * ((c/12.92) if (c<=0.04045) else pow((c+0.055)/1.055, 2.4))
import sys, math
sys.stdout.write("#include <stdint.h>\n")
if False: # 1D illumination invariant image
sys.stdout.write("const uint8_t invariant_table[65536] = {\n") # 65536 * 1
for i in range(65536):
r = ((((i >> 3) & 31) * 255) + 15.5) // 31
g = (((((i & 7) << 3) | (i >> 13)) * 255) + 31.5) // 63
b = ((((i >> 8) & 31) * 255) + 15.5) // 31
# http://ai.stanford.edu/~alireza/publication/cic15.pdf
r_lin = lin(r / 255.0) + 1.0
g_lin = lin(g / 255.0) + 1.0
b_lin = lin(b / 255.0) + 1.0
r_lin_sharp = (r_lin * 0.9968) + (g_lin * 0.0228) + (b_lin * 0.0015);
g_lin_sharp = (r_lin * -0.0071) + (g_lin * 0.9933) + (b_lin * 0.0146);
b_lin_sharp = (r_lin * 0.0103) + (g_lin * -0.0161) + (b_lin * 0.9839);
lin_sharp_avg = r_lin_sharp * g_lin_sharp * b_lin_sharp
lin_sharp_avg_p = math.pow(lin_sharp_avg, 1.0/3.0) if (lin_sharp_avg > 0.0) else 0.0
r_lin_sharp_div = 0.0
g_lin_sharp_div = 0.0
b_lin_sharp_div = 0.0
if lin_sharp_avg_p > 0.0:
lin_sharp_avg_d = 1.0 / lin_sharp_avg_p
r_lin_sharp_div = r_lin_sharp * lin_sharp_avg_d
g_lin_sharp_div = g_lin_sharp * lin_sharp_avg_d
b_lin_sharp_div = b_lin_sharp * lin_sharp_avg_d
r_lin_sharp_div_log = math.log(r_lin_sharp_div) if (r_lin_sharp_div > 0.0) else 0.0
g_lin_sharp_div_log = math.log(g_lin_sharp_div) if (g_lin_sharp_div > 0.0) else 0.0
b_lin_sharp_div_log = math.log(b_lin_sharp_div) if (b_lin_sharp_div > 0.0) else 0.0
chi_x = (r_lin_sharp_div_log * 0.7071) + (g_lin_sharp_div_log * -0.7071) + (b_lin_sharp_div_log * 0.0000)
chi_y = (r_lin_sharp_div_log * 0.4082) + (g_lin_sharp_div_log * 0.4082) + (b_lin_sharp_div_log * -0.8164)
chi_int = max(min(int(round(math.exp((chi_x * 0.9326) + (chi_y * -0.3609)) * 127.5)), 255), 0)
if not (i % 16):
sys.stdout.write(" ")
sys.stdout.write("%3d" % chi_int)
if (i + 1) % 16:
sys.stdout.write(", ")
elif i != 65535:
sys.stdout.write(",\n")
else:
sys.stdout.write("\n};\n")
if True: # 2D illumination invariant image
sys.stdout.write("const uint16_t invariant_table[65536] = {\n") # 65536 * 2
for i in range(65536):
r = ((((i >> 3) & 31) * 255) + 15.5) // 31
g = (((((i & 7) << 3) | (i >> 13)) * 255) + 31.5) // 63
b = ((((i >> 8) & 31) * 255) + 15.5) // 31
# http://ai.stanford.edu/~alireza/publication/cic15.pdf
r_lin = lin(r / 255.0) + 1.0
g_lin = lin(g / 255.0) + 1.0
b_lin = lin(b / 255.0) + 1.0
r_lin_sharp = (r_lin * 0.9968) + (g_lin * 0.0228) + (b_lin * 0.0015);
g_lin_sharp = (r_lin * -0.0071) + (g_lin * 0.9933) + (b_lin * 0.0146);
b_lin_sharp = (r_lin * 0.0103) + (g_lin * -0.0161) + (b_lin * 0.9839);
lin_sharp_avg = r_lin_sharp * g_lin_sharp * b_lin_sharp
lin_sharp_avg_p = math.pow(lin_sharp_avg, 1.0/3.0) if (lin_sharp_avg > 0.0) else 0.0
r_lin_sharp_div = 0.0
g_lin_sharp_div = 0.0
b_lin_sharp_div = 0.0
if lin_sharp_avg_p > 0.0:
lin_sharp_avg_d = 1.0 / lin_sharp_avg_p
r_lin_sharp_div = r_lin_sharp * lin_sharp_avg_d
g_lin_sharp_div = g_lin_sharp * lin_sharp_avg_d
b_lin_sharp_div = b_lin_sharp * lin_sharp_avg_d
r_lin_sharp_div_log = math.log(r_lin_sharp_div) if (r_lin_sharp_div > 0.0) else 0.0
g_lin_sharp_div_log = math.log(g_lin_sharp_div) if (g_lin_sharp_div > 0.0) else 0.0
b_lin_sharp_div_log = math.log(b_lin_sharp_div) if (b_lin_sharp_div > 0.0) else 0.0
chi_x = (r_lin_sharp_div_log * 0.7071) + (g_lin_sharp_div_log * -0.7071) + (b_lin_sharp_div_log * 0.0000)
chi_y = (r_lin_sharp_div_log * 0.4082) + (g_lin_sharp_div_log * 0.4082) + (b_lin_sharp_div_log * -0.8164)
e_t_x = 0.9326
e_t_y = -0.3609
p_th_00 = e_t_x * e_t_x
p_th_01 = e_t_x * e_t_y
p_th_10 = e_t_y * e_t_x
p_th_11 = e_t_y * e_t_y
x_th_x = (p_th_00 * chi_x) + (p_th_01 * chi_y)
x_th_y = (p_th_10 * chi_x) + (p_th_11 * chi_y)
r_chi = (x_th_x * 0.7071) + (x_th_y * 0.4082)
g_chi = (x_th_x * -0.7071) + (x_th_y * 0.4082)
b_chi = (x_th_x * 0.0000) + (x_th_y * -0.8164)
r_chi_invariant = math.exp(r_chi)
g_chi_invariant = math.exp(g_chi)
b_chi_invariant = math.exp(b_chi)
chi_invariant_sum = r_chi_invariant + g_chi_invariant + b_chi_invariant
r_chi_invariant_m = 0.0
g_chi_invariant_m = 0.0
b_chi_invariant_m = 0.0
if chi_invariant_sum > 0.0:
chi_invariant_sum = 1.0 / chi_invariant_sum
r_chi_invariant_m = r_chi_invariant * chi_invariant_sum
g_chi_invariant_m = g_chi_invariant * chi_invariant_sum
b_chi_invariant_m = b_chi_invariant * chi_invariant_sum
rgb565 = (int((((r_chi_invariant_m*255)*31)+127.5)/255)&0x1F)<<11 | \
(int((((g_chi_invariant_m*255)*63)+127.5)/255)&0x3F)<<5 | \
(int((((b_chi_invariant_m*255)*31)+127.5)/255)&0x1F)
if not (i % 16):
sys.stdout.write(" ")
sys.stdout.write("0x%04X" % (((rgb565&0x00ff)<<8)|((rgb565&0xff00)>>8)))
if (i + 1) % 16:
sys.stdout.write(", ")
elif i != 65535:
sys.stdout.write(",\n")
else:
sys.stdout.write("\n};\n")