Compiled Cython Code - Detects colors in images 2-3 x faster than Numpy

pip install locate-pixelcolor-cythonsingle

Tested+compiled against Windows 10 / Python 3.10 / Anaconda

If you can't import it, compile it on your system (code at the end of this page)

How to use it in Python

import numpy as np
import cv2
from locate_pixelcolor_cythonsingle import search_colors
# 4525 x 6623 x 3 picture https://www.pexels.com/pt-br/foto/foto-da-raposa-sentada-no-chao-2295744/
picx = r"C:\Users\hansc\Downloads\pexels-alex-andrews-2295744.jpg"
pic = cv2.imread(picx)
colors0 = np.array([[255, 255, 255]],dtype=np.uint8)
resus0 = search_colors(pic=pic, colors=colors0)
colors1=np.array([(66,  71,  69),(62,  67,  65),(144, 155, 153),(52,  57,  55),(127, 138, 136),(53,  58,  56),(51,  56,  54),(32,  27,  18),(24,  17,   8),],dtype=np.uint8)
resus1 =  search_colors(pic=pic, colors=colors1)
####################################################################
%timeit resus0 = search_colors(pic=pic, colors=colors0)
51 ms ± 201 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

b,g,r = pic[...,0],pic[...,1],pic[...,2]
%timeit np.where(((b==255)&(g==255)&(r==255)))
150 ms ± 209 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
####################################################################
%timeit resus1 =  search_colors(pic=pic, colors=colors1)
443 ms ± 1.19 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit np.where(((b==66)&(g==71)&(r==69))|((b==62)&(g==67)&(r==65))|((b==144)&(g==155)&(r==153))|((b==52)&(g==57)&(r==55))|((b==127)&(g==138)&(r==136))|((b==53)&(g==58)&(r==56))|((b==51)&(g==56)&(r==54))|((b==32)&(g==27)&(r==18))|((b==24)&(g==17)&(r==8)))
1 s ± 16.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
####################################################################

The Cython Code

# cython: language_level=3

import numpy as np
cimport numpy as np

cpdef searchforcolor(np.uint8_t[::1] pic, np.uint8_t[::1] colors, int width, int totallengthpic, int totallengthcolor, int[::1] outputx, int[ ::1] outputy, int[::1] lastresult):
    cdef int counter = 0
    cdef unsigned char r, g, b
    cdef int i, j

    for i in range(0, totallengthcolor, 3):
        r = colors[i]
        g = colors[i + 1]
        b = colors[i + 2]
        for j in range(0, totallengthpic, 3):
            if ( r== pic[j]) and (g == pic[j+1]) and (b == pic[j+2]):
                dividend = j // 3
                quotient = dividend // width
                remainder = dividend % width
                outputx[counter] = quotient
                outputy[counter] = remainder
                lastresult[0] = counter
                counter += 1

# .\python.exe .\colorsearchcythonsinglesetup.py build_ext --inplace

setup.py to compile the code

# cython: language_level=3

from setuptools import Extension, setup
from Cython.Build import cythonize
import numpy as np
ext_modules = [
    Extension("colorsearchcythonsingle", ["colorsearchcythonsingle.pyx"], include_dirs=[np.get_include()],define_macros=[("NPY_NO_DEPRECATED_API", "NPY_1_7_API_VERSION")])
]

setup(
    name='colorsearchcythonsingle',
    ext_modules=cythonize(ext_modules),
)


# .\python.exe .\colorsearchcythonsinglesetup.py build_ext --inplace

Alternatives

I wrote a couple of variations of this function. All of them can be used in Python.

Cython, but with multiple processors (5-10x faster than Numpy)

https://github.com/hansalemaos/locate_pixelcolor_cythonmulti

Cupy, using the GPU (up to 8x faster than Numpy)

https://github.com/hansalemaos/locate_pixelcolor_cupy

C - shared library (10x faster than Numpy)

https://github.com/hansalemaos/locate_pixelcolor_c

C++ - parallel_for - shared library (up to 10x faster than Numpy)

https://github.com/hansalemaos/locate_pixelcolor_cpp_parallelfor

C++ - pragma omp - shared library (20x faster than Numpy)

https://github.com/hansalemaos/locate_pixelcolor_cpppragma

Numba - compiled - ahead of time (2-3x faster than numpy)

https://github.com/hansalemaos/locate_pixelcolor_numba

Numba Cuda - compiled - ahead of time (10x faster than numpy)

https://github.com/hansalemaos/locate_pixelcolor_numbacuda