Home / Questions / 226799
Accepted
Jesus_jbs
Asked: 2020-01-03 10:22:13 +0800 CST 2020-01-03 10:22:13 +0800 CST

How to modify color intensities in an image with numpy. python

  • 772

I have the following definition to create a function that displays an image in grayscale:

def escala_grises(imagen):
''' Transforma una imagen en color a grados de grises

ENTRADA: 
   - imagen: matriz de píxeles -> [[(float, float, float)]]
SALIDA: 
   - matriz de píxeles con la imagen de entrada en escala de grises -> [[float]]

Toma como entrada una matriz de píxeles RGB, y produce como salida una
matriz de píxeles con intensidades en escala de grises (un único valor por
cada pixel).

Para convertir una imagen en color a escala de grises hay que aplicar
los siguientes pesos para integrar los tres canales de color en un único
valor de intensidad:
    - Rojo:  0.2989
    - Verde: 0.5870
    - Azul:  0.1140
'''

The argument imagenis an array extracted from a ".jpeg" file using an array. Ex: [[[82, 119, 171], [91, 128, 180], [85, 122, 174], [84, 121, 173], [88, 125, 177], [84, 121, 173], [84, 121, 173], [85, 122, 174]]].

I'm guessing it's a 3D array since the method shapereturns (330, 600, 3), the 3 I think comes from the number of colors present in the "RGB" file.

The question is, how to modify the array so that the average color intensities match [0.2989, 0.5870, 0.1140], considering that I can only use the following imported modules:

from matplotlib import pyplot as plt
import numpy as np

-Reply to abulafia regarding the comments Apparently, I shouldn't use métodosthe library numpysince I'm not supposed to have given them. So I tried to recreate what you have told me using the following code: 

filtro = []
for i in imagen:
    for r,g,b in i:
        rojo = r*0.2989
        verde = g*0.5870
        azul = b*0.1140
        numero = [int(rojo + verde + azul)]
        filtro.append(numero)
return filtro

But it gives a blank image with the error:

ValueError: Image size of 1x198000 pixels is too large. It must be less than 2^16 in each direction.

Attached Test of the grayscale function and the function to display the image:

def test_escala_grises(imagen):
    grises = escala_grises(imagen)
    muestra_imagen(grises, cmap='gray')
    guarda_imagen('../img/grises.jpeg', grises, cmap='gray')

def muestra_imagen(imagen, cmap=None):
    imagen = np.array(imagen, dtype='uint8')
    plt.imshow(imagen, cmap=cmap)
    plt.show()
python

1 Answers

  1. Best Answer

    The code that converts each pixel to grayscale is correct. But you have an error:

    • The list that you return at the end of your function escala_grises()is no longer a two-dimensional array, like the input image was, but a mere sequence of pixels. You must use reshapeto return it to its original two-dimensional shape, otherwise plt.imgshow()it won't know how to display it.
    • You should use the cmap="Gray_r"to use a grayscale where 0 represents black and 255 represents white (the default is the other way around).

    Fixing these issues:

    import numpy as np
import matplotlib.pyplot as plt

def escala_grises(imagen):
  filtro = []
  for i in imagen:
      for r,g,b in i:
          rojo = r*0.2989
          verde = g*0.5870
          azul = b*0.1140
          numero = [int(rojo + verde + azul)]
          filtro.append(numero)
  # Usamos los valores en imagen.shape para conocer el ancho x alto originales
  # y reformateamos la lista para que use esas dimensiones
  result = np.reshape(filtro, imagen.shape[:2])
  return result


def muestra_imagen(imagen, cmap="Greys_r"):
    # Con el nuevo código no es necesario convertirlo a array numpy
    # pues la función escala_grises ya retorna uno
    # imagen = np.array(imagen, dtype='uint8')
    plt.imshow(imagen, cmap=cmap)
    plt.show()

    Example of use (with the following image, taken from here )

    input image

    im = plt.imread("no-bonsai-kitten.jpg", format="jpeg")
grises = escala_grises(im)
muestra_imagen(grises)

    Result:

    Result

    Update

    So that it is not necessary to use np.reshape(), the result returned by escala_grises()should be a list of lists, so that each of its elements represents a row (with as many elements as there were columns in the original image).

    The following code implements this version:

    def escala_grises(imagen):
  result = []  # Esta lista será de "filas"
  for fila in range(imagen.shape[0]):
    aux = []   # Creamos una fila vacia
    for columna in range(imagen.shape[1]):
      pixel = imagen[fila][columna]
      gris = pixel[0]*0.2989 + pixel[1]*0.5870 + pixel[2]*0.1140
      # Añadimos este pixel a la fila
      aux.append(int(gris))
    # Una vez terminada la fila, la añadimos a la lista general
    result.append(aux)
  return result
    • 1