先用灰度平均值二值化,然后在图片上纵向找出每一行的分割点使图片两边的差异最大,然后把它们连起来成为一条直线

import cv2
import numpy as np

# 读取图像并转换为灰度图
image = cv2.imread('1.jpg', cv2.IMREAD_GRAYSCALE)

# 计算灰度图的平均值
mean_gray_value = np.mean(image)

# 使用平均灰度值作为阈值进行二值化
_, binary = cv2.threshold(image, mean_gray_value, 255, cv2.THRESH_BINARY)

# 获取图像的高度和宽度
height, width = binary.shape

# 存储分割点的坐标
split_points = []

# 定义一个函数来计算给定分割点的左右侧差异
def calculate_variance_difference(row, x):
    left_side = row[:x]
    right_side = row[x:]
    if len(left_side) == 0 or len(right_side) == 0:
        return 0
    variance_left = np.count_nonzero(left_side)/len(left_side)
    variance_right = np.count_nonzero(right_side)/len(right_side)
    return abs(variance_left - variance_right)

# 遍历每一行，寻找最佳分割点
for y in range(height):
    row = binary[y, :]
    best_x = None
    max_diff = 0
    for x in range(width//2-100, width//2+100): ### 根据实际情况调整范围!!
        diff = calculate_variance_difference(row, x)
        if diff > max_diff:
            max_diff = diff
            best_x = x
    if best_x is not None:
        split_points.append((best_x, y))

# 将分割点转换为NumPy数组
split_points_np = np.array(split_points)

# 拟合直线
if len(split_points_np) >= 2:  # 至少需要两个点才能拟合一条直线
    [vx, vy, x0, y0] = cv2.fitLine(split_points_np, cv2.DIST_L2, 0, 0.01, 0.01)
    lefty = int((-x0*vy/vx) + y0)
    righty = int(((width-x0)*vy/vx) + y0)
else:
    print("无法拟合直线，因为分割点不足")
    exit()

# 绘制直线
color_image = cv2.cvtColor(binary, cv2.COLOR_GRAY2BGR)  # 转换为彩色以便绘制不同颜色的线
cv2.line(color_image, (0, lefty), (width-1, righty), (0, 255, 0), 2)

# 显示结果
cv2.imshow('Image with Split Line', color_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

# 如果需要保存分割后的图像
cv2.imwrite('image_with_split_line.jpg', color_image)