import cv2

1. Image Transform

This work tried to transform images from one color-space to another. The transformation you can see below:

• mask
• segment
• deskew
• gray
• thresh
• rnoise
• canny
• sharpen

According to [2] while the color image can be treated arbitrary vector value functions or collections of independent bands, it usually makes sense to think about them as highly correlated signals with strong connections to the image formation process, sensor design, and Human perception. Consider example brightening picture by adding a constant value to all three channels. In fact, adding the same value to each color channel not only increases the apparent intensity of each pixel, but it cal also affects the picture hue and saturation.

In [7]:

link code

#masking function
def create_mask_for_image(image):
    image_hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)

    lower_hsv = np.array([0,0,250])
    upper_hsv = np.array([250,255,255])

    mask = cv2.inRange(image_hsv, lower_hsv, upper_hsv)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (11,11))
    mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
    return mask

#image  deskew function
def  deskew_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image  gray  function
def  gray_image(image):
    mask = create_mask_for_image(image)
    output = cv2.cvtColor(image,  cv2.COLOR_BGR2GRAY)
    return output/255

#image  thresh  function
def  thresh_image(image):
    img = read_img(df['file'][250],(255,255))
    gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    output = cv2.threshold(gray,0,255,cv2.THRESH_BINARY_INV) #+cv.THRESH_OTSU)
    return output

#image  rnoise  function
def  rnoise_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image  dilate  function
def  dilate_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image  erode  function
def  erode_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image  opening  function
def  opening_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image canny function
def  canny_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#image segmentation function
def segment_image(image):
    mask = create_mask_for_image(image)
    output = cv2.bitwise_and(image, image, mask = mask)
    return output/255

#sharpen the image
def sharpen_image(image):
    image_blurred = cv2.GaussianBlur(image, (0, 0), 3)
    image_sharp = cv2.addWeighted(image, 1.5, image_blurred, -0.5, 0)
    return image_sharp

# function to get an image
def read_img(filepath, size):
    img = image.load_img(os.path.join(data_kaggle, filepath), target_size=size)
    #convert image to array
    img = image.img_to_array(img)
    return img</pre>

SHOW SAMPLE IMAGES

nb_rows = 3
nb_cols = 5
fig, axs = plt.subplots(nb_rows, nb_cols, figsize=(10, 5));   #打印成3行5列
plt.suptitle('SAMPLE IMAGES');
for i in range(0, nb_rows):
    for j in range(0, nb_cols):
        axs[i, j].xaxis.set_ticklabels([]);
        axs[i, j].yaxis.set_ticklabels([]);
        axs[i, j].imshow((read_img(df['file'][np.random.randint(400)], (255,255)))/255.);
plt.show();

打印一行两列

#get an image
img = read_img(df['file'][12],(255,255))

#mask
image_mask = create_mask_for_image(img)

fig, ax = plt.subplots(1, 2, figsize=(5, 5));
plt.suptitle('RESULT', x=0.5, y=0.8)
plt.tight_layout(1)

ax[0].set_title('ORIGINAL', fontsize=12)
ax[1].set_title('MASK', fontsize=12)

ax[0].imshow(img/255);
ax[1].imshow(image_mask);

原文
https://www.kaggle.com/code/khotijahs1/cv-image-preprocessing
https://www.kaggle.com/code/jamesmcguigan/image-preprocessing