Contents

Color transform

2. Color transform

import matplotlib.pyplot as plt
import cv2

package

rgb <-> bgr

rgb to gray

gray to rgb

cv2

import cv2
img_bgr = cv2.imread(“xx.jpg”)
cv_rgb = cv2.cvtColor(cv_bgr, cv2.COLOR_BGR2RGB)
cv_bgr = cv2.cvtColor(cv_rgb, cv2.COLOR_RGB2BGR)

gray_img = cv2.cvtColor(cv_rgb, cv2.COLOR_RGB2GRAY)
gray_img = cv2.cvtColor(cv_bgr, cv2.COLOR_BGR2GRAY)

# formula: gray = (np.round(0.2989xR + 0.5870xG + 0.1140xB)).astype("uint8")
# data type: uint8; shape: (height, width)

gray_to_rgb = cv2.cvtColor(gray_img, cv2.COLOR_GRAY2BGR)

# fromula: [gray_img, gray_img, gray_img]
# data type: uint8; shape: (height, width, 3)

skimage

import matplotlib.pyplot as plt
from skimage import color

img_rgb = plt.imread(“xx.jpg”)
img_gray = color.rgb2gray(rocket_image)
# data type: uint8; shape: (height, width)

plt

import matplotlib.pyplot as plt
import numpy as np

img_rgb = plt.imread(“xx.jpg”)
img_bgr = img_rgb[:,:,::-1]
img_rgb = img_bgr[:,:,::-1]

2.1. RGB 與 BGR 轉換

cv_bgr = cv2.imread("imgs/fruits-2.jpg")

plt.imshow(cv_bgr);
_images/color_transform_4_0.png

  • 可以看到，這張圖跟我想的不一樣，藍色的地方，本來應該是紅色的。這是因為讀進來時，預設是 BGR 的通道

  • 我們可以透過 cv2 的 cvtColor() 來轉換，例如：

cv_rgb = cv2.cvtColor(cv_bgr, cv2.COLOR_BGR2RGB)
plt.imshow(cv_rgb);
_images/color_transform_6_0.png

  • 也可以轉回來

cv_bgr = cv2.cvtColor(cv_rgb, cv2.COLOR_RGB2BGR)
plt.imshow(cv_bgr);
_images/color_transform_8_0.png

  • 也可以不要用 function，就用 array 的特性來改就好

cv_rgb = cv_bgr[:,:,::-1]
plt.imshow(cv_rgb);
_images/color_transform_10_0.png 
cv_bgr = cv_rgb[:,:,::-1]
plt.imshow(cv_bgr);
_images/color_transform_11_0.png

2.2. 彩色轉灰階

  • cv2 的彩色轉灰階的公式為： gray = (np.round(0.2989xR + 0.5870xG + 0.1140xB)).astype("uint8")

  • 可以看到，他給三個通道不同的權重，以G通道最多，再來是R，最後是B。而且算完後，會先四捨五入，再轉成 uint8 (0~255無號整數)

  • 用 cv2 的 function，就可以把 RGB轉灰階，或 BGR 轉灰階，如下：

gray_img = cv2.cvtColor(cv_rgb, cv2.COLOR_RGB2GRAY)
print(f"The data type of image is : {gray_img.dtype}")
print(f"The shape of image is : {gray_img.shape}")
plt.imshow(gray_img, cmap = "gray");

The data type of image is : uint8
The shape of image is : (417, 626)
_images/color_transform_14_1.png

  • 可以看到，轉成灰階後，通道被拿掉了

  • 再來看看，BGR 轉灰階，作法一樣：

gray_img = cv2.cvtColor(cv_bgr, cv2.COLOR_BGR2GRAY)
print(f"The data type of image is : {gray_img.dtype}")
print(f"The shape of image is : {gray_img.shape}")
plt.imshow(gray_img, cmap = "gray");

The data type of image is : uint8
The shape of image is : (417, 626)
_images/color_transform_17_1.png

  • 自己用公式來轉也可以

def my_rgb_to_gray(rgb_img):
    r = rgb_img[:,:,0]
    g = rgb_img[:,:,1]
    b = rgb_img[:,:,2]
    gray = 0.2989*r + 0.5870*g + 0.1140*b
    gray = np.round(gray)
    return gray.astype("uint8")

gray_img2 = my_rgb_to_gray(cv_rgb)
print(f"The data type of image is : {gray_img2.dtype}")
print(f"The shape of image is : {gray_img2.shape}")
plt.imshow(gray_img2, cmap = "gray");

The data type of image is : uint8
The shape of image is : (417, 626)
_images/color_transform_19_1.png

  • 如果用 skimage，就這樣做：

import matplotlib.pyplot as plt
from skimage import color

img_rgb = plt.imread("imgs/fruits-2.jpg")
img_gray = color.rgb2gray(img_rgb)

print(f"The data type of image is : {gray_img.dtype}")
print(f"The shape of image is : {img_gray.shape}")
plt.imshow(img_gray, cmap = "gray");

The data type of image is : uint8
The shape of image is : (417, 626)
_images/color_transform_21_1.png

  • 可以看到，shape也是把通道拿掉了。但轉換公式沒查，不確定是不是和 cv2 一樣

2.3. 灰階轉彩色(新增通道)

  • open cv 的轉換公式為：color image = [gray_img, gray_img, gray_img]

  • 所以，就是灰階影像，幫你重複三次

  • 要注意的是，原本的 gray_img 要是 uint8 的 data type，如果不是，要先幫他轉一下

gray_to_rgb = cv2.cvtColor(gray_img, cv2.COLOR_GRAY2BGR)
print(f"The shape of image is : {gray_to_rgb.shape}")
plt.imshow(gray_to_rgb);

The shape of image is : (417, 626, 3)
_images/color_transform_25_1.png

2.4. 其他應用

  • opencv 的書 (CH4-3~4-7) 還有很多內容，這等之後再整理了

    • 4-3: HSV 色彩空間, BGR/RGB <-> HSV

    • 4-4: 拆分 channel

    • 4-5: 合併 channel

    • 4-6: 拆分與合併 channel 的應用 (色調 Hue 調整, 飽和 Saturation 調整, 明度 Value 調整)

    • 4-7: alpha 通道

previous

1. Read, show, and save imgs