如何删除触及图像边界的对象（单元格）？

2024-05-20 • 问答

我有一张细胞图像，我对它进行了阈值处理，并检测到了这些细胞（使用 cv2）。

我想创建一个值为 True 或 False 的数组，以显示每个组件是否接触到图像的边界 (True) 或不接触 (False)。

import cv2 as cv

# Read the image you want connected components of,IN BLACK AND WHITE
img = cv.imread('../images/37983_ERSyto/cellpaintingfollowup-reimage_a01_s1_w26ae36209-938b-45ef-b166-3aba3af125df.tif',cv.IMREAD_GRAYSCALE)

seed_pt = (100,800) #point in the background
fill_color = 0
mask = np.zeros_like(img)
kernel = cv.getStructuringElement(cv.MORPH_RECT,(6,5))
for th in range(7,70):
    #creates a copy of the mask:
    prev_mask = mask.copy()
    #thresholded image:
    mask = cv.threshold(img,th,22331,cv.THRESH_BINARY)[1]
    #FloodFill: fill a connected component starting from the seed point with the specified color.
    mask = cv.floodFill(mask,None,seed_pt,fill_color)[1]
    #cv.bitwise: calculates the per-element bit-wise disjunction of two arrays or an array and a scalar. Superposition of thresholded images
    mask = cv.bitwise_or(mask,prev_mask)
    #clean speckles
    mask = cv.morphologyEx(mask,cv.MORPH_OPEN,kernel)
   

#compute the connected components labeled image of boolean image and also produce a statistics output for each label

connectivity = 8  #You need to choose 4 or 8 for connectivity type.

#OBTAIN FEATURE OF THE AREA IN PIXELS OF THE CELLS
stats = cv.connectedComponentsWithStats(mask,connectivity,cv.CV_32S)[2]
label_area = stats[1:,cv.CC_STAT_AREA] #we dont include the first element because it represents the area of the background

#OBTAIN FEATURES OF THE CENTROID POSITION
centroids = cv.connectedComponentsWithStats(mask,cv.CV_32S)[3]
label_centroids_x = centroids[1:,0] #dont include the first element because it represents the background
label_centroids_y = centroids[1:,1]

#HIGHT: The vertical size of the bounding box.
label_hight = stats[1:,cv.CC_STAT_HEIGHT]
#WIDTH: The horizontal size of the bounding box.
label_width = stats[1:,cv.CC_STAT_WIDTH]


#TOUCHING IMAGE BOUNDARIES: is the component touching the boundaries of the matrix/image?--> True/False

label_bounary = #boolean array

我首先想到的是搜索每个组件的轮廓并定义一些限制，但是我无法理解每个组件的标签是如何存储的，因此我无法选择所需的组件。

这是图片：

非常感谢您。

使用您的代码（感谢您的评论），我得到了这个面具。它可能不一样，因为 .jpg 压缩可能会弄乱图像（这不是无损压缩方案）

@fmw42 完全正确，他在我完成代码之前评论了

import cv2 as cv
import numpy as np

# Read the image you want connected components of,IN BLACK AND WHITE
img = cv.imread('cells.jpg',cv.IMREAD_GRAYSCALE)

seed_pt = (100,800) #point in the background
fill_color = 0
mask = np.zeros_like(img)
kernel = cv.getStructuringElement(cv.MORPH_RECT,(6,5))
for th in range(7,70):
    #creates a copy of the mask:
    prev_mask = mask.copy()
    #thresholded image:
    mask = cv.threshold(img,th,22331,cv.THRESH_BINARY)[1]
    #FloodFill: fill a connected component starting from the seed point with the specified color.
    mask = cv.floodFill(mask,None,seed_pt,fill_color)[1]
    #cv.bitwise: calculates the per-element bit-wise disjunction of two arrays or an array and a scalar. Superposition of thresholded images
    mask = cv.bitwise_or(mask,prev_mask)
    #clean speckles
    mask = cv.morphologyEx(mask,cv.MORPH_OPEN,kernel)

# show mask
cv.imshow("Mask",mask);
cv.waitKey(0);

# contours OpenCV 3.4,if you're using OpenCV 2 or 4,it returns (contours,_)
_,contours,_ = cv.findContours(mask,cv.RETR_TREE,cv.CHAIN_APPROX_NONE);

# get bounds and check if they're touching edge
height,width = mask.shape[:2];
touching_edge = []; # boolean array,index matches the contours list
for con in contours:
    # get bounds
    x,y,w,h = cv.boundingRect(con);

    # check if touching edge
    on_edge = False;
    if x <= 0 or (x + w) >= (width - 1):
        on_edge = True;
    if y <= 0 or (y + h) >= (height - 1):
        on_edge = True;

    # add to list
    touching_edge.append(on_edge);

# mark the contours on the edge
colored = cv.cvtColor(mask,cv.COLOR_GRAY2BGR);
for index in range(len(contours)):
    if touching_edge[index]:
        # drawContours(image,contour_list,index,color,thickness) # -1 is filled
        cv.drawContours(colored,(50,50,200),-1);

# show
cv.imshow("Colored",colored);
cv.waitKey(0);

如果您愿意使用 scikit-image，可以尝试使用 clear_border：

>>> import numpy as np
>>> from skimage.segmentation import clear_border
>>> labels = np.array([[0,1,0],...                    [1,...                    [0,0]])
>>> clear_border(labels)
array([[0,[0,0]])

如何删除触及图像边界的对象（单元格）？

iCMS 回答：如何删除触及图像边界的对象（单元格）？

大家都在问