Python 3.x 将HSV遮罩拆分为多个矩形

我从图像中创建了一个HSV遮罩。结果如下：

我的目标是绘制适合遮罩高度或宽度的多重矩形，如下所示：

我遇到了2道题

我不知道如何在创建矩形的掩码中定位起点和终点。若我使用for循环逐行扫描遮罩，它可能会将遮罩分成两部分。

有时，在一张图像中还包含两个不同的遮罩。如何绘制矩形？

---

任何人都可以给我一些建议吗？

您可以使用

cv2.findContour（）

搜索您的最大轮廓（十字形状）。它返回轮廓的坐标数组。然后，您可以搜索具有最高X坐标（即最右点）、最低X坐标（最左点）、最高Y坐标（即最底点）和最低Y坐标（即最高点）的轮廓点。在获得所有4个值后，可以再次搜索轮廓中具有该值的所有点，并将它们附加到4个不同的列表中，稍后对它们进行排序，以便将这些点作为淹没在底图上：

从这一点开始，我将列表转换为numpy数组，因为这对我来说更容易。你可以用任何其他方法

然后，问题就在于你想要多少个矩形，以及你想如何显示它们。在我的示例代码中，您必须输入需要多少相同大小的矩形，最后一个是剩余矩形的大小。我首先在Y坐标（绿色）上显示矩形，然后在X坐标上显示矩形，X坐标分为两段（左侧和右侧），因为它们的距离略有不同，我不想在Y坐标矩形上绘制，因为它们没有在示例图像上绘制。您可以根据需要更改写入矩形的逻辑。希望它能对你有所帮助，或者给你一个如何处理的想法。干杯

示例代码：

import cv2
import numpy as np

# Read image and search for contours. 
img = cv2.imread('cross.png')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, threshold = cv2.threshold(gray, 100, 255, cv2.THRESH_BINARY)
_, contours, hierarchy = cv2.findContours(threshold,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)

# Select the biggest contour (if you wish to segmentize only the cross-like contour). 
cnt = max(contours, key=cv2.contourArea)

# Create empty lists for appending key points. 
top_vertical = []
bottom_vertical = []
left_horizontal = []
right_horizontal = []

# Setting the starter values for N, S, E, W. 
top = 10000
bottom = 0
left = 10000
right = 0

# Loop to get highest key values of N, S, E, W.  
for i in cnt:
    y = int(i[:,1])
    x = int(i[:, 0])
    if x < left:
        left = int(x)
    if x > right:
        right = int(x)
    if y < top:
        top = int(y)
    if y > bottom:
        bottom = int(y)

# Loop for appending all points containing key values of N, S, E, W.   
for i in cnt:
    if int(i[:,1]) == top:
        up = (int(i[:,0]), int(i[:,1]))
        top_vertical.append(up)
    if int(i[:,1]) == bottom:
        down = (int(i[:,0]), int(i[:,1]))
        bottom_vertical.append(down)
    if int(i[:,0]) == left:
        l = (int(i[:,0]), int(i[:,1]))
        left_horizontal.append(l)
    if int(i[:,0]) == right:
        r = (int(i[:,0]), int(i[:,1]))
        right_horizontal.append(r)


# Sorting the lists. 
top_vertical.sort(key=lambda tup: tup[0])
bottom_vertical.sort(key=lambda tup: tup[0])
left_horizontal.sort(key=lambda tup: tup[1])
right_horizontal.sort(key=lambda tup: tup[1])

# Optional drawing of key points. 
'''cv2.circle(img,(top_vertical[0]), 4, (0,0,255), -1)
cv2.circle(img,(top_vertical[-1]), 4, (0,0,255), -1)
cv2.circle(img,(bottom_vertical[0]), 4, (0,0,255), -1)
cv2.circle(img,(bottom_vertical[-1]), 4, (0,0,255), -1)
cv2.circle(img,(left_horizontal[0]), 4, (0,0,255), -1)
cv2.circle(img,(left_horizontal[-1]), 4, (0,0,255), -1)
cv2.circle(img,(right_horizontal[0]), 4, (0,0,255), -1)
cv2.circle(img,(right_horizontal[-1]), 4, (0,0,255), -1)'''

# Transforming lists to arrays. 
top_vertical = np.array(top_vertical)
bottom_vertical = np.array(bottom_vertical)
left_horizontal = np.array(left_horizontal)
right_horizontal = np.array(right_horizontal)

# Calculating height and weight of the contour.
distance_y = bottom - top
distance_x = right - left

# Inputs for the number of same size segments. 
a = input('Input the number of same size segments in Y coordinate: ')
b = input('Input the number of same size segments in left X coordinate: ')
c = input('Input the number of same size segments in right X coordinate: ')

# Calculation of area per segment and limit for the lenght of combined segments (height and weight) . 
segment_y = distance_y/int(a)
segment_x_reference = int(top_vertical[0,0]) - int(left_horizontal[0,0])
segment_x = segment_x_reference/int(b)
segment_x_right_reference = int(right_horizontal[0,0]) - int(top_vertical[-1,0])
segment_x_right = segment_x_right_reference/int(c)

# Drawing rectangles on the Y axis.
for i in range(1,20):
    sq = int(segment_y)*i
    if sq < distance_y:
        cv2.rectangle(img,(top_vertical[0,0], top_vertical[0,1]),((top_vertical[-1,0]),top_vertical[0,1] + sq),(0,255,0),1)
    else:
        sq = distance_y
        cv2.rectangle(img,(top_vertical[0,0], top_vertical[0,1]),((top_vertical[-1,0]),sq),(0,255,0),1)
        break

# Drawing rectangles on the left side of X axis.   
for i in range(1,20):
    sq = int(segment_x)*i
    if sq < segment_x_reference:
        cv2.rectangle(img,(left_horizontal[0,0], left_horizontal[0,1]),((left_horizontal[0,0])+sq, left_horizontal[-1,1]),(255,0,0),1)
    else:
        sq = segment_x_reference
        cv2.rectangle(img,(left_horizontal[0,0], left_horizontal[0,1]),((left_horizontal[0,0])+sq, left_horizontal[-1,1]),(255,0,0),1)
        break

# Drawing rectangles on the right side of X axis. 
for i in range(1,20):
    sq = int(segment_x_right)*i
    if sq < segment_x_right_reference:
        cv2.rectangle(img,(right_horizontal[0,0], right_horizontal[0,1]),((right_horizontal[0,0])-sq, right_horizontal[-1,1]),(255,0,0),1)
    else:
        sq = segment_x_right_reference
        cv2.rectangle(img,(right_horizontal[0,0], right_horizontal[0,1]),((right_horizontal[0,0])-sq, right_horizontal[-1,1]),(255,0,0),1)
        break

# Displaying result. 
cv2.imshow('img', img)

---

您的矩形是否有特定的尺寸需要考虑，或者每个遮罩中是否有最佳数量的矩形需要绘制？激发灵感的帖子+1。这种方法也适用于不规则形状的轮廓吗？谢谢！我认为这是行不通的，因为它假设外墙是直线，并且它有一个十字架形状。如果你改变了极值点的阈值，如果它有多个交叉点（如t形）或没有交叉点（如普通正方形）等，它可能适用于不规则形状。但这可能需要大量工作：）很好的解决方案。但是，如果交叉形状围绕图像中的某个噪声点，则可能会影响4个值（上、右、左、下）的提取。你有什么办法解决这个问题吗？嗯……如果只有一两个像素，你可以将“if int（i[：，1]）==top”改为“if int（i[：，1]）>=top-2”等等，或者你可以制作附加列表来附加top、left、right、bottom的值，并选择具有最高值且在列表中出现一次或两次以上的值。

import cv2
import numpy as np

# Read image and search for contours. 
img = cv2.imread('cross.png')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, threshold = cv2.threshold(gray, 100, 255, cv2.THRESH_BINARY)
_, contours, hierarchy = cv2.findContours(threshold,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)

# Select the biggest contour (if you wish to segmentize only the cross-like contour). 
cnt = max(contours, key=cv2.contourArea)

# Create empty lists for appending key points. 
top_vertical = []
bottom_vertical = []
left_horizontal = []
right_horizontal = []

# Setting the starter values for N, S, E, W. 
top = 10000
bottom = 0
left = 10000
right = 0

# Loop to get highest key values of N, S, E, W.  
for i in cnt:
    y = int(i[:,1])
    x = int(i[:, 0])
    if x < left:
        left = int(x)
    if x > right:
        right = int(x)
    if y < top:
        top = int(y)
    if y > bottom:
        bottom = int(y)

# Loop for appending all points containing key values of N, S, E, W.   
for i in cnt:
    if int(i[:,1]) == top:
        up = (int(i[:,0]), int(i[:,1]))
        top_vertical.append(up)
    if int(i[:,1]) == bottom:
        down = (int(i[:,0]), int(i[:,1]))
        bottom_vertical.append(down)
    if int(i[:,0]) == left:
        l = (int(i[:,0]), int(i[:,1]))
        left_horizontal.append(l)
    if int(i[:,0]) == right:
        r = (int(i[:,0]), int(i[:,1]))
        right_horizontal.append(r)


# Sorting the lists. 
top_vertical.sort(key=lambda tup: tup[0])
bottom_vertical.sort(key=lambda tup: tup[0])
left_horizontal.sort(key=lambda tup: tup[1])
right_horizontal.sort(key=lambda tup: tup[1])

# Optional drawing of key points. 
'''cv2.circle(img,(top_vertical[0]), 4, (0,0,255), -1)
cv2.circle(img,(top_vertical[-1]), 4, (0,0,255), -1)
cv2.circle(img,(bottom_vertical[0]), 4, (0,0,255), -1)
cv2.circle(img,(bottom_vertical[-1]), 4, (0,0,255), -1)
cv2.circle(img,(left_horizontal[0]), 4, (0,0,255), -1)
cv2.circle(img,(left_horizontal[-1]), 4, (0,0,255), -1)
cv2.circle(img,(right_horizontal[0]), 4, (0,0,255), -1)
cv2.circle(img,(right_horizontal[-1]), 4, (0,0,255), -1)'''

# Transforming lists to arrays. 
top_vertical = np.array(top_vertical)
bottom_vertical = np.array(bottom_vertical)
left_horizontal = np.array(left_horizontal)
right_horizontal = np.array(right_horizontal)

# Calculating height and weight of the contour.
distance_y = bottom - top
distance_x = right - left

# Inputs for the number of same size segments. 
a = input('Input the number of same size segments in Y coordinate: ')
b = input('Input the number of same size segments in left X coordinate: ')
c = input('Input the number of same size segments in right X coordinate: ')

# Calculation of area per segment and limit for the lenght of combined segments (height and weight) . 
segment_y = distance_y/int(a)
segment_x_reference = int(top_vertical[0,0]) - int(left_horizontal[0,0])
segment_x = segment_x_reference/int(b)
segment_x_right_reference = int(right_horizontal[0,0]) - int(top_vertical[-1,0])
segment_x_right = segment_x_right_reference/int(c)

# Drawing rectangles on the Y axis.
for i in range(1,20):
    sq = int(segment_y)*i
    if sq < distance_y:
        cv2.rectangle(img,(top_vertical[0,0], top_vertical[0,1]),((top_vertical[-1,0]),top_vertical[0,1] + sq),(0,255,0),1)
    else:
        sq = distance_y
        cv2.rectangle(img,(top_vertical[0,0], top_vertical[0,1]),((top_vertical[-1,0]),sq),(0,255,0),1)
        break

# Drawing rectangles on the left side of X axis.   
for i in range(1,20):
    sq = int(segment_x)*i
    if sq < segment_x_reference:
        cv2.rectangle(img,(left_horizontal[0,0], left_horizontal[0,1]),((left_horizontal[0,0])+sq, left_horizontal[-1,1]),(255,0,0),1)
    else:
        sq = segment_x_reference
        cv2.rectangle(img,(left_horizontal[0,0], left_horizontal[0,1]),((left_horizontal[0,0])+sq, left_horizontal[-1,1]),(255,0,0),1)
        break

# Drawing rectangles on the right side of X axis. 
for i in range(1,20):
    sq = int(segment_x_right)*i
    if sq < segment_x_right_reference:
        cv2.rectangle(img,(right_horizontal[0,0], right_horizontal[0,1]),((right_horizontal[0,0])-sq, right_horizontal[-1,1]),(255,0,0),1)
    else:
        sq = segment_x_right_reference
        cv2.rectangle(img,(right_horizontal[0,0], right_horizontal[0,1]),((right_horizontal[0,0])-sq, right_horizontal[-1,1]),(255,0,0),1)
        break

# Displaying result. 
cv2.imshow('img', img)

Input the number of same size segments in Y coordinate: 5
Input the number of same size segments in left X coordinate: 2
Input the number of same size segments in right X coordinate: 2