在 python 中取 OpenCV 的最大連接分量

Question

我有一些帶有 alpha 通道的葉子照片 - 背景 alpha 通道為 0，照片中的葉子像素 alpha 通道為 255。

我想將圖像分離為連接的組件。 然后取大連接組件。 為了像紅色標記一樣擦除污垢。 這是我在這里談論的圖像：葉子圖像

這是我當前的代碼：

def remove_pixels(image_path, img_file_output_path):
    img = cv2.imread(image_path, cv2.IMREAD_UNCHANGED)
    alpha = img[:, :, 3]
    alpha = cv2.threshold(alpha, 0, 255, cv2.THRESH_BINARY)[1]
    num_labels, labels_im = cv2.connectedComponents(alpha,connectivity=4)
    new_img = np.zeros(img.shape)
    largest_label = 1 + np.argmax(labels_im[1:, cv2.CC_STAT_AREA])  
    h, w = labels_im.shape
    for i in range(h*w):
        x = i // w
        y = i % w
        if labels_im[x,y] == largest_label:
            new_img[x,y] = img[x,y]
    cv2.imwrite(img_file_output_path, new_img)

它不起作用，非常感謝任何幫助。
謝謝：）
-------------------------------------------------

這是我的新代碼：

def remove_pixels(image_path, img_file_output_path):
    img = cv2.imread(image_path, cv2.IMREAD_UNCHANGED)
    alpha = img[:, :, 3]
    contours = cv2.findContours(alpha, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    contours = contours[0] if len(contours) == 2 else contours[1]
    big_contour = max(contours, key=cv2.contourArea)

    # draw white filled contour on black background to use as new alpha channel
    new_alpha = np.zeros_like(alpha)
    new_alpha = cv2.drawContours(new_alpha, [big_contour], 0, 255, -1)

    # put new_alpha into alpha channel of img
    new_img = cv2.cvtColor(img, cv2.COLOR_BGR2BGRA)
    new_img[:, :, 3] = new_alpha

    cv2.imwrite(img_file_output_path, new_img)

和圖片：新圖片

Answer 1

這是 Python/OpenCV 中的一種方法。

 - Read the input
 - Threshold on white and invert so that the leaf, etc is white on black background
 - Get the contours
 - Filter the contours on area to keep only the largest
 - Draw a white filled contour on a black background using the largest contour
 - Put that image into the alpha channel of the input
 - Save the result

輸入：

import cv2
import numpy as np

# load image
img = cv2.imread('leaf3.jpg')

# threshold on white and invert
thresh = cv2.inRange(img, (220,220,220), (255,255,255))
thresh = 255 - thresh

# get the largest contour
contours = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]
big_contour = max(contours, key=cv2.contourArea)

# draw white filled contour on black background to use as new alpha channel
new_alpha = np.zeros_like(thresh)
new_alpha = cv2.drawContours(new_alpha, [big_contour], 0, 255, -1)

# put new_alpha into alpha channel of img 
new_img = cv2.cvtColor(img, cv2.COLOR_BGR2BGRA)
new_img[:, :, 3] = new_alpha

# save result
cv2.imwrite("leaf3_alpha_cleaned.png", new_img)

結果：

可選 - 將背景設置為白色而不是透明。

# make background white
new_img2 = img.copy()
new_img2[new_alpha==0] = (255,255,255)
cv2.imwrite("leaf3_alpha_cleaned2.png", new_img2)