收集所有不同轮廓的非零像素

Question

我一直在尝试根据一个序列对轮廓进行排序（序列在这里并不重要）。我有一个非常小的问题，那就是我应该在下面的代码片段中传递的正确的numpy数组，我可以同时获得正确的行/列（非零）像素。你知道吗

row_pixels=cv2.countNonZero(blur[cy][:])

col_pixels=cv2.countNonZero(blur[:,cx])

我所做的是得到如下结果：对于所有的5个轮廓，我得到几乎相同数量的非零像素，我意识到这是因为我传递了“整个”图像（正如你看到上面的模糊是整个图像）作为numpy数组来计算像素，这是错误的，我意识到。你知道吗

当前输入：下图（无标记）

预期输出：对于所有5个轮廓，行/列（非零）像素。你知道吗

我现在做的是：

import cv2
import numpy as np
from imutils import perspective
from imutils import contours 
import imutils 
from scipy.spatial import distance as dist
import argparse
import pandas as pd
import time

parser = argparse.ArgumentParser(description='Object Detection and Tracking using YOLO in OPENCV')
parser.add_argument('--image', help='Path to image file.')

args = parser.parse_args()
font=cv2.FONT_HERSHEY_SIMPLEX


start=time.time()
im_in = cv2.imread(args.image, 0)
_, thres2=cv2.threshold(im_in, 140, 255,cv2.THRESH_BINARY_INV)
dilate = cv2.dilate(thres2,None)
erode = cv2.erode(dilate,None)

im_3=erode.copy()

blur=cv2.medianBlur(im_3,5)

a=[]
r=[] 
row_col_pixel_values=[]
cl=[]
data=[]
global mainar
#find contours 
_,contour2,_=cv2.findContours(blur,cv2.RETR_CCOMP,cv2.CHAIN_APPROX_NONE)
# print(contour2)
for c in contour2:
    area=cv2.contourArea(c)
    if area>10000 and area <30000:
        a.append(area)

        cv2.drawContours(blur, [c], 0, (128, 128, 128), 1)


        M=cv2.moments(c)
        cx=int((M["m10"]/M["m00"]))
        cy=int((M["m01"]/M["m00"]))
        center =(cx,cy)
        data.append((cx,cy))
        cv2.circle(blur,(cx,cy), 5,(128,128,128),-1)
        print("",cx,cy)
        print(len(blur[cy][:])) 
        # one=blur[c]
        row_pixels=cv2.countNonZero(blur[cy][:])

        col_pixels=cv2.countNonZero(blur[:,cx])
        comb=(row_pixels,col_pixels)
        cl.append(comb)

nparea=np.array(a)
npcentercoord=np.array(data)

row_col_pixel_values=np.array(cl)
print("Area of 5 contours :",nparea)
print("Center coordinates of 5 contours:",npcentercoord)

print("Row and Column pixel values of 5 contours:",row_col_pixel_values)

mainar=np.column_stack((nparea,npcentercoord,row_col_pixel_values))
# print(mainar)

mainar[:,[1]] = (mainar[:,[1]]).astype(int)

MinX = int(min([_[1] for _ in mainar]))
MinlowerX = (MinX - 10) 
MinupperX = (MinX + 10)
MinY = int(min([_[2] for _ in mainar]))
MinlowerY = (MinY - 10) 
MinupperY = (MinY + 10)
MaxX = int(max([_[1] for _ in mainar]))
MaxlowerX = (MaxX - 10) 
MaxupperX = (MaxX + 10)
MaxY = int(max([_[2] for _ in mainar]))
MaxlowerY = (MaxY - 10)
MaxupperY = (MaxY + 10)

print("", MinX,MinY,MaxX,MaxY)


def PixeltoNumeric(channel,rowMM,colMM):

    if channel=="4S":
        for i in range(0, len(mainar[:,1])):
            cx=mainar[i,1]
            cy=mainar[i,2]
            if (cx in range(MinlowerX,MinupperX+1)) and (cy in range(MinlowerY,MinupperY+1)):
                rowp=mainar[i,3]
                colp=mainar[i,4]
                print("The center coordinates(x,y) and (Row/Col) pixels of 4Schannel: ")
                print("1 pixel has {:.5f} many mm in row:".format(rowMM/rowp))
                print("1 pixel has {:.5f} many mm in col:".format(colMM/colp))
                print(cx,cy,rowp,colp)

    if channel == '1':
            for i in range(0, len(mainar[:,1])):
                cx=mainar[i,1]
                cy=mainar[i,2]
                if (cx in range(MaxlowerX,MaxupperX+1)) and (cy in range(MaxlowerY,MaxupperY+1)):
                    rowp=mainar[i,3]
                    colp=mainar[i,4]
                    print("The center coordinates(x,y) and (Row/Col) pixels of 1Channel: ")
                    print("1 pixel has {:.5f} many mm in row:".format(rowMM/rowp))
                    print("1 pixel has {:.5f} many mm in col:".format(colMM/colp))
                    print(cx,cy,rowp,colp)

    if channel == '2':
        for i in range(0, len(mainar[:,1])):
            cx=mainar[i,1]
            cy=mainar[i,2]
            if (cx in range(MinlowerX,MinupperX+1)) and (cy in range(MaxlowerY,MaxupperY+1)):
                rowp=mainar[i,3]
                colp=mainar[i,4]
                print("The center coordinates(x,y) and (Row/Col) pixels of 2Channel: ")
                print("1 pixel has {:.5f} many mm in row:".format(rowMM/rowp))
                print("1 pixel has {:.5f} many mm in col:".format(colMM/colp))
                print(cx,cy,rowp,colp)

    if channel == '3':
        for i in range(0, len(mainar[:,1])):
            cx=mainar[i,1]
            cy=mainar[i,2]
            if (cx in range(((MinlowerX+MaxlowerX)//2),((MinupperX+MaxupperX+1)//2)) and (cy in range(((MinlowerY+MaxlowerY)//2),((MinupperY+MaxupperY+1)//2)))):
                rowp=mainar[i,3]
                colp=mainar[i,4]
                print("The center coordinates(x,y) and (Row/Col) pixels of 3Channel: ")
                print("1 pixel has {:.5f} many mm in row:".format(rowMM/rowp))
                print("1 pixel has {:.5f} many mm in col:".format(colMM/colp))
                print(cx,cy,rowp,colp)

    if channel == '4N':
        for i in range(0, len(mainar[:,1])):
            cx=mainar[i,1]
            cy=mainar[i,2]
            if (cx in range(MaxlowerX,MaxupperX+1)) and (cy in range(MinlowerY,MinupperY+1)):
                rowp=mainar[i,3]
                colp=mainar[i,4]
                print("The center coordinates(x,y) and (Row/Col) pixels of 4NChannel: ")
                print("1 pixel has {:.5f} many mm in row:".format(rowMM/rowp))
                print("1 pixel has {:.5f} many mm in col:".format(colMM/colp))
                print(cx,cy,rowp,colp)

    return (cv2.imshow("4",blur))


cv2.waitKey(0)
cv2.destroyAllWindows()

Answer 1

我不完全确定我是否正确理解你，但我的理解是，你想找到所有轮廓内像素的所有坐标（x，y）。如果这是您的问题，您可以通过以下代码实现：

import cv2
import matplotlib.pyplot as plt
import numpy as np

im_in = cv2.imread(r'image.png', 0)
_, thres2 = cv2.threshold(im_in, 140, 255, cv2.THRESH_BINARY_INV)
dilate = cv2.dilate(thres2, None)
erode = cv2.erode(dilate, None)
im_3 = erode.copy()
blur = cv2.medianBlur(im_3, 5)

# I am using OpenCV 4 therefore it returns only 4 parameters
contour2, _ = cv2.findContours(blur, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_NONE)
extracted = np.zeros(blur.shape, np.uint8)

for c in contour2:
    area = cv2.contourArea(c)
    # I have modified these values to make it work for attached picture
    if 10000 < area < 300000: 
        cv2.drawContours(extracted, [c], 0, (255), cv2.FILLED)

contour_x, contour_y = np.nonzero(extracted)

plt.imshow(extracted, 'gray')
plt.show()

这是提取的image。你知道吗

更新1

经过你的解释，我知道你想计算每个单独轮廓的宽度和高度。根据您提供的示例代码，我假设您希望使用与轮廓中心相交的线来测量宽度和高度。您可以通过在清晰的图像上绘制和测量轮廓来实现这一点。请参见下面的代码：

# I am using OpenCV 4 therefore it returns only 4 parameters
contour2, _ = cv2.findContours(blur, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_NONE)
extracted = np.zeros(blur.shape, np.uint8)
contoursSize = []
for c in contour2:
    area = cv2.contourArea(c)
    # I have modified these values to make it work for attached picture
    if 10000 < area < 300000:
        M = cv2.moments(c)
        cx = int((M["m10"] / M["m00"]))
        cy = int((M["m01"] / M["m00"]))
        extracted.fill(0) 
        cv2.drawContours(extracted, [c], 0, 255, cv2.FILLED)
        width = cv2.countNonZero(extracted[cy][:])
        height = cv2.countNonZero(extracted[:, cx])
        contoursSize.append((width, height))