如何在 opencv python 中获得低对比度图像的边缘

Question

I'm trying to get the edges of this object from a TEM(microscope) image and the problem is that the contact is low especially in the upper edge, I tried several things thresholding, contrast equalization... but I wasn't able to get the upper edge.

NB: I'm trying to calculate the angle between the droplet and the tube I'm not sure if this is the best way to approach this problem.

The original image:

The Canny Edge detection I get:

the steps I got to get this result are:

1. Contrast enhancement
2. Thresholding
3. Gauss filter
4. Canny Edge detection

Code:

clahe = cv2.createCLAHE(clipLimit=clip_limit, tileGridSize=(grid_size, grid_size))
equ = clahe.apply(img)
val = filters.threshold_otsu(equ)
mask = img < val
# denoising part
mask = filters.gaussian(mask,sigma=sigmaG)
# edge detection
edge = feature.canny(mask,sigma=sigmaC)
edge = img_as_ubyte(edge)

Answer 1

We have this image and we want to detect the edges of the microphone:

Basically, I converted the image to grayscale, added a Gaussian blur, and detected the edges using the canny edge detector. One more important part is to fill in the gaps in the detected edges by dilating the edges and then eroding them.

All of the above is implemented in the process function; the draw_contours function basically utilizes the process function, and detects the greatest contour:

import cv2
import numpy as np

def process(img):
    img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    img_blur = cv2.GaussianBlur(img_gray, (11, 11), 7)
    img_canny = cv2.Canny(img_blur, 0, 42)
    kernel = np.ones((19, 19))
    img_dilate = cv2.dilate(img_canny, kernel, iterations=4)
    img_erode = cv2.erode(img_dilate, kernel, iterations=4)
    return img_erode

def draw_contours(img):
    contours, hierarchies = cv2.findContours(process(img), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    cnt = max(contours, key=cv2.contourArea)
    peri = cv2.arcLength(cnt, True)
    approx = cv2.approxPolyDP(cnt, 0.004 * peri, True)
    cv2.drawContours(img, [approx], -1, (255, 255, 0), 2)

img = cv2.imread("image.jpg")
h, w, c = img.shape

img = cv2.resize(img, (w // 2, h // 2))
draw_contours(img)

cv2.imshow("Image", img)
cv2.waitKey(0)

Output:

You can omit the drop by tweaking some values int the process function. For example, the values

def process(img):
    img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    img_blur = cv2.GaussianBlur(img_gray, (11, 11), 10)
    img_canny = cv2.Canny(img_blur, 0, 38)
    kernel = np.ones((13, 13))
    img_dilate = cv2.dilate(img_canny, kernel, iterations=3)
    img_erode = cv2.erode(img_dilate, kernel, iterations=4)
    return img_erode

Output: