Improve border of the result of watershed

Question

I tried to count the leaves of this image by taking the code of this post:

enter link description here

Some things were changed, like cleaning the image with morphological operations

I want to separate the shapes that look like two united circles with a line, since I believe the algorithm would work better. Besides, I have a doubt regarding the final result, because the lines are not closed. How can I improve the code or the image for the final result and have the lines closed like the ones in the first image?

import cv2 as cv
import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import label
import urllib.request


# https://stackoverflow.com/a/14617359/7690982


def function(a, img, img_gray):

    # Added several elliptical structuring element for better morphology process
    struct_big = cv.getStructuringElement(cv.MORPH_ELLIPSE,(4,4))
    struct_small = cv.getStructuringElement(cv.MORPH_ELLIPSE,(3,3))

    # increase border size
    border = cv.dilate(img, struct_big, iterations=1)
    border = border - cv.erode(img, struct_small)

    dt = cv.distanceTransform(img, cv.DIST_L2, 3)
    dt = ((dt - dt.min()) / (dt.max() - dt.min()) * 255).astype(np.uint8)


    # blur the signal lighty to remove noise
    dt = cv.GaussianBlur(dt,(7,7),-1)

    # Adaptive threshold to extract local maxima of distance trasnform signal
    
    #PARA LA PRIMERA IMAGEN 11 -5
    #35 45 o 50 -5
    #13, -5
    dt = cv.adaptiveThreshold(dt, 255, cv.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY, 11, -5)

    # Morphology operation to clean the thresholded signal
    dt = cv.erode(dt,struct_small,iterations = 1)
    dt = cv.dilate(dt,struct_big,iterations = 4)

    # Labeling
    lbl, ncc = label(dt)
    lbl = lbl * (255 / (ncc + 1))
    print(lbl)
    # Completing the markers now.
    lbl[border >= 80] = 255
    lbl = lbl.astype(np.int32)
    cv.watershed(a, lbl)
    print("[INFO] {} unique segments found".format(len(np.unique(lbl)) - 1))
    lbl[lbl == -1] = 0
    lbl = lbl.astype(np.uint8)
    return 255 - lbl

if __name__ == '__main__':    

    img = cv.imread('/home/rainor/Escritorio/VA/Practica2/PSI_Tray031_2015-12-27--09-03-57_top.png')
    # img = cv.imread('/home/rainor/Escritorio/VA/Practica2/PSI_Tray032_2016-01-14--14-06-50_top.png')
    # Open Image

    kernel = np.array([[-1,-1,-1], [-1,9,-1], [-1,-1,-1]])
    img = cv.filter2D(img, -1, kernel)
    plt.imshow(img)
    plt.show()

    # proper color segmentation
    hsv = cv.cvtColor(img, cv.COLOR_BGR2HSV)  
    mask = cv.inRange(hsv, (25, 25, 25), (70, 255, 255))

    mask = cv.medianBlur(mask, 9)
    kernel = np.ones((12,12),np.uint8)
    mask = cv.morphologyEx(mask, cv.MORPH_OPEN, kernel)

    print("Esta es la que quitamos el ruido")
    plt.imshow(mask, cmap='gray')
    plt.show()


    imask = mask > 0
    slicer = np.zeros_like(img, np.uint8)
    slicer[imask] = img[imask]

    # Image Binarization
    img_gray = cv.cvtColor(slicer, cv.COLOR_BGR2GRAY)

    print("Blanco y negro: check")
    plt.imshow(img_gray, cmap = 'gray')
    plt.show()

    _, img_bin = cv.threshold(img_gray, 0, 100, cv.THRESH_BINARY)

    print("Imagen plantas Amarillas fondo morado (tiene suciedda igual quitando como dije en BN el ruido)")
    plt.imshow(img_bin, cmap='gray')
    plt.show()

    # Morphological Gradient
    cv.morphologyEx(img_bin, cv.MORPH_OPEN,cv.getStructuringElement(cv.MORPH_ELLIPSE,(1,1)),img_bin,(-1,-1),10)
    cv.morphologyEx(img_bin, cv.MORPH_ERODE,cv.getStructuringElement(cv.MORPH_ELLIPSE,(1,1)),img_bin,(-1,-1),3)

    # Segmentation
    result = function(img, img_bin, img_gray)
    plt.imshow(np.hstack([result, img_gray]), cmap='Set3')
    plt.show()

    # Final Picture
    result[result != 255] = 0
    result = cv.dilate(result, None)
    img[result == 255] = (0, 0, 255)
    plt.imshow(result)
    plt.show()

    # function(img1)

Original images

https://imgur.com/a/v7wY3ET

Result of border image 1

Final result of image 1

Result of border image 2

Final result of image 1