这是一个简单但非常有趣的颜色分割练习。这个主题已经在各处得到了广泛的报道,有几个例子遍布Stackoverflow。在许多情况下,颜色分割在 HSV 颜色空间中效果最好。
在左下图中,你可以看到带有蓝色孔洞的黄砖的分割结果,只是为了表明它们也被这种方法检测到。
在这个答案中,我提供了检测黄砖并识别其中的孔所需的操作的高级概述。但是,它没有演示如何计算特定砖块内的孔数以避免破坏您的作业。我故意省略了答案的那部分,以便为您留下一些工作。
以下是我的方法的主要步骤:
- 然后对分割后的图像进行处理,我们丢弃小块以仅保留最大的块(即砖块)。有了这个过滤机制,就可以统计有多少黄砖了。这里有一个绝妙的技巧:如果您使用
cv2.fillPoly() 绘制砖的轮廓并用白色填充,您将能够在单独的图像中绘制没有任何孔的整个砖来创建蒙版。这很快就会派上用场!黄色面具如下所示:
- 在这个阶段,我们已经有了图像中所有黄砖的位置。剩下要做的就是识别每块砖上的洞。这就是蒙版的用武之地:如果你注意上面的两张图片,分割后的图像和蒙版的区别主要是砖块的洞:
- 处理此图像的轮廓允许丢弃所有不属于孔的小斑点,只留下砖的孔。我们可以在分割图像或原始图像上绘制孔的位置以显示它们:
总之,这段代码提供了一个黄砖列表和另一个包含这些砖中孔的列表。从这一点开始,由你决定。该代码可以轻松扩展以处理其他颜色的积木。玩得开心:
import cv2
import numpy as np
# convertToOpenCVHSV():
# converts from HSV range (H: 0-360, S: 0-100, V: 0-100)
# to what OpenCV expects: (H: 0-179, S: 0-255, V: 0-255)
def convertToOpenCVHSV(H, S, V):
return np.array([H // 2, S * 2.55, V * 2.55], np.uint8)
# 1. Load input image
img = cv2.imread('test_images/legos.jpg')
# 2. Preprocess: quantize the image to reduce the number of colors
div = 6
img = img // div * div + div // 2
cv2.imwrite('lego2_quantized.jpg', img)
# 3. Convert to HSV color space
hsv_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
# 4. Segment the image using predefined values of yellow (min and max colors)
low_yellow = convertToOpenCVHSV(40, 35, 52)
high_yellow = convertToOpenCVHSV(56, 95, 93)
yellow_seg_img = cv2.inRange(hsv_img, low_yellow, high_yellow)
#cv2.imshow('yellow_seg_img', yellow_seg_img)
cv2.imwrite('lego4_yellow_seg_img.jpg', yellow_seg_img)
# 5. Identify and count the number of yellow bricks and create a mask with just the yellow objects
bricks_list = []
min_size = 5
contours, hierarchy = cv2.findContours(yellow_seg_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
for contourIdx, cnt in enumerate(contours):
# filter out tiny segments
x, y, w, h = cv2.boundingRect(cnt)
if (w < min_size) or (h < min_size):
continue
#print('contourIdx=', contourIdx, 'w=', w, 'h=', h)
bricks_list.append(cnt)
# debug: draw green contour in the original image
#cv2.drawContours(img, cnt, -1, (0, 255, 0), 2) # green
print('Detected', len(bricks_list), 'yellow pieces.')
# Iterate the list of bricks and draw them (filled) on a new image to be used as a mask
yellow_mask_img = np.zeros((img.shape[0], img.shape[1]), np.uint8)
for cnt in bricks_list:
cv2.fillPoly(yellow_mask_img, pts=[cnt], color=(255,255,255))
cv2.imshow('yellow_mask_img', yellow_mask_img)
cv2.imwrite('lego5_yellow_mask_img.jpg', yellow_mask_img)
# debug: display only the original yellow bricks found
bricks_img = cv2.bitwise_and(img, img, mask=yellow_mask_img)
#cv2.imshow('bricks_img', bricks_img)
cv2.imwrite('lego5_bricks_img.jpg', bricks_img)
# 6. Identify holes in each Lego brick
diff_img = yellow_mask_img - yellow_seg_img
cv2.imshow('diff_img', diff_img)
cv2.imwrite('lego6_diff_img.jpg', diff_img)
# debug: create new BGR image for debugging purposes
dbg_img = cv2.cvtColor(yellow_mask_img, cv2.COLOR_GRAY2RGB)
#dbg_img = bricks_img
holes_list = []
min_area_size = 10
max_area_size = 24
contours, hierarchy = cv2.findContours(yellow_seg_img, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
for contourIdx, cnt in enumerate(contours):
# filter out tiny segments by area
area = cv2.contourArea(contours[contourIdx])
if (area < min_area_size) or (area > max_area_size):
#print('contourIdx=', contourIdx, 'w=', w, 'h=', h, 'area=', area, '(ignored)')
#cv2.drawContours(dbg_img, cnt, -1, (0, 0, 255), 2) # red
continue
#print('contourIdx=', contourIdx, 'w=', w, 'h=', h, 'area=', area)
holes_list.append(cnt)
# debug: draw a blue-ish contour on any BGR image to show the holes of the bricks
for cnt in holes_list:
cv2.fillPoly(dbg_img, pts=[cnt], color=(255, 128, 0))
cv2.fillPoly(img, pts=[cnt], color=(255, 128, 0))
cv2.imwrite('lego6_dbg_img.jpg', dbg_img)
cv2.imwrite('lego6_img.jpg', img)
# 7. Iterate though the list of holes and associate them with a particular brick
# TODO
cv2.imshow('img', img)
cv2.imshow('dbg_img', dbg_img)
cv2.waitKey(0)