在此图像中找到毛巾最低角(而不是边缘)的最佳方法(使用python和OpenCV)是什么?
同样,毛巾的颜色可以不同,但​​是背景颜色将始终相同。

python - OpenCV检测毛巾边距-LMLPHP

我需要这个角落(最低的“真实”毛巾角落):
python - OpenCV检测毛巾边距-LMLPHP

最佳答案

由于图像中大约有两种不同的颜色(前景和背景各有一种),因此可以将图像转换为HSV颜色空间并可视化每个单独的通道。
代码:

path = r'C:\Users\Desktop'
filename = 'towel.jpg'

img = cv2.imread(os.path.join(path, filename))
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)     #--- convert to HSV
cv2.imshow('hsv.jpg', hsv)
h = hsv[:,:,0]
cv2.imshow('h.jpg', h)                         #--- visualize the hue channel
python - OpenCV检测毛巾边距-LMLPHP
ret, thresh = cv2.threshold(h, 0, 255, cv2.THRESH_OTSU + cv2.THRESH_BINARY_INV)
cv2.imshow('thresh1', thresh)                  #--- apply Otsu threshold on hue channel
python - OpenCV检测毛巾边距-LMLPHP
请注意,毛巾中央的白色斑点必须去除。为此,我使用了形态学开口法。
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(25, 25))
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel)
cv2.imshow('fin', cv2.bitwise_not(opening))
python - OpenCV检测毛巾边距-LMLPHP
编辑
OpenCV提供了查找给定轮廓的顶部,底部,最右边和最左边的角的功能。我获得了最终结果图像的轮廓,并找到了四个极端。
代码:
im2, contours, hierarchy = cv2.findContours(cv2.bitwise_not(opening), cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)                #--- finding contours
cnt = contours[0]                                 #--- since there is only one contour present

leftmost = tuple(cnt[cnt[:,:,0].argmin()][0])
rightmost = tuple(cnt[cnt[:,:,0].argmax()][0])
topmost = tuple(cnt[cnt[:,:,1].argmin()][0])
bottommost = tuple(cnt[cnt[:,:,1].argmax()][0])

print('The extreme points are leftmost: {}, rightmost: {}, topmost: {} and bottommost: {}'.format(leftmost, rightmost, topmost, bottommost))

The extreme points are leftmost: (32, 336), rightmost: (807, 439), topmost: (459, 12) and bottommost: (699, 743)
我还在原始图像的副本上标出了极端点:
img2 = img.copy()
cv2.circle(img2, leftmost, 5, (0, 255, 255), -1)    #-- leftmost
cv2.circle(img2, rightmost, 5, (0, 255, 255), -1)    #-- rightmost
cv2.circle(img2, topmost, 5, (0, 255, 255), -1)    #-- topmost
cv2.circle(img2, bottommost, 5, (0, 255, 255), -1)    #-- bottommost
python - OpenCV检测毛巾边距-LMLPHP

10-02 13:46