opencv 轮廓合并

最新推荐文章于 2024-06-15 17:37:41 发布

Java阿忠

最新推荐文章于 2024-06-15 17:37:41 发布

阅读量414

点赞数 4

分类专栏： opencv 文章标签： opencv 人工智能计算机视觉

本文链接：https://blog.csdn.net/weixin_45664361/article/details/138345579

版权

opencv 专栏收录该内容

1 篇文章 0 订阅

订阅专栏

opencv 轮廓合并

本文将指导你将多个轮廓合并为一个轮廓，主要操作逻辑为：寻找各轮廓之间的最短距离，并按最短距离对应的点进行轮廓拼接，以达到轮廓合并的效果。
以下为运算处理的结果
示例对照

实现步骤

opencv的导入

pip install opencv-python

或者

pip install opencv-python-headless

加载图像与轮廓

加载包括多个轮廓的图像

check_piece_image = cv2.imread("check_piece_image.jpg")
cv2.imshow("check_piece_image", check_piece_image)

check_piece_image
做灰度图处理

check_piece_gray_image = cv2.cvtColor(check_piece_image, cv2.COLOR_BGR2GRAY)
cv2.imshow("check_piece_gray_image", check_piece_gray_image)

check_piece_gray_image
因为图像是白底黑字，则需要按位取反，如果你们不是，那么这步可以忽略

check_piece_bitwise_not_image = cv2.bitwise_not(check_piece_gray_image)
cv2.imshow("check_piece_bitwise_not_image", check_piece_bitwise_not_image)

check_piece_bitwise_not_image
阙值化处理

# _, check_piece_thresh_image = cv2.threshold(check_piece_gray_image, 100, 255, cv2.THRESH_OTSU)
_, check_piece_thresh_image = cv2.threshold(check_piece_bitwise_not_image, 100, 255, cv2.THRESH_OTSU)
cv2.imshow("check_piece_thresh_image", check_piece_thresh_image)

check_piece_thresh_image
寻找所有的轮廓

check_piece_contours, _ = cv2.findContours(check_piece_thresh_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
check_piece_contours_image = check_piece_image.copy()
cv2.drawContours(check_piece_contours_image, check_piece_contours, -1, (0, 255, 0), 2)
cv2.imshow("check_piece_contours_image", check_piece_contours_image)

check_piece_contours_image

对轮廓进行合并

合并轮廓寻找最小距离并合并为一个轮廓

new_merged_check_piece_contours = merge_contours(check_piece_contours)
check_piece_contours_image_by_merged = check_piece_image.copy()
cv2.drawContours(check_piece_contours_image_by_merged, new_merged_check_piece_contours, -1, (0, 255, 0), 2)
cv2.imshow("check_piece_contours_image_by_merged", check_piece_contours_image_by_merged)

核心方法

def merge_contours(contours: typing.Sequence[cv2.typing.MatLike]) -> typing.Sequence[cv2.typing.MatLike]:
    """合并轮廓 寻找最小距离并合并为一个轮廓

    Args:
        contours (typing.Sequence[cv2.typing.MatLike]): 轮廓

    Returns:
        typing.Sequence[cv2.typing.MatLike]: 合并后的轮廓
    """
    contours = list(contours).copy()
    if len(contours) <= 1:
        return contours
    min_distance = float('inf')
    closest_i = None
    closest_j = None
    new_contour = None
    for i in range(len(contours)):
        for j in range(i + 1, len(contours)):
            candidate_contour, distance = merge_two_contour_to_one(contours[i], contours[j])
            if distance < min_distance:
                new_contour, min_distance, closest_i, closest_j = candidate_contour, distance, i, j
    contours[closest_i] = new_contour
    del contours[closest_j]
    return merge_contours(contours)

def merge_two_contour_to_one(contour1: cv2.typing.MatLike, contour2: cv2.typing.MatLike) -> tuple[cv2.typing.MatLike, float]:
    min_distance, closest_i, closest_j = float('inf'), None, None
    for i, point1 in enumerate(contour1):
        for j, point2 in enumerate(contour2):
            distance = np.linalg.norm(point1[0] - point2[0], 2, None, False)
            if distance < min_distance:
                min_distance, closest_i, closest_j = distance, i, j
    rearranged_contour1 = np.vstack((contour1[closest_i:], contour1[:closest_i]))
    rearranged_contour2 = np.vstack((contour2[closest_j:], contour2[:closest_j]))
    return np.vstack((rearranged_contour1, rearranged_contour2)), min_distance

check_piece_contours_image_by_merged

全部代码

所有代码以及所有过程截图

import cv2
from cv2.typing import MatLike
import typing
import numpy as np

def merge_contours(contours: typing.Sequence[cv2.typing.MatLike]) -> typing.Sequence[cv2.typing.MatLike]:
    """合并轮廓 寻找最小距离并合并为一个轮廓

    Args:
        contours (typing.Sequence[cv2.typing.MatLike]): 轮廓

    Returns:
        typing.Sequence[cv2.typing.MatLike]: 合并后的轮廓
    """
    contours = list(contours).copy()
    if len(contours) <= 1:
        return contours
    min_distance = float('inf')
    closest_i = None
    closest_j = None
    new_contour = None
    for i in range(len(contours)):
        for j in range(i + 1, len(contours)):
            candidate_contour, distance = merge_two_contour_to_one(contours[i], contours[j])
            if distance < min_distance:
                new_contour, min_distance, closest_i, closest_j = candidate_contour, distance, i, j
    contours[closest_i] = new_contour
    del contours[closest_j]
    return merge_contours(contours)

def merge_two_contour_to_one(contour1: cv2.typing.MatLike, contour2: cv2.typing.MatLike) -> tuple[cv2.typing.MatLike, float]:
    min_distance, closest_i, closest_j = float('inf'), None, None
    for i, point1 in enumerate(contour1):
        for j, point2 in enumerate(contour2):
            distance = np.linalg.norm(point1[0] - point2[0], 2, None, False)
            if distance < min_distance:
                min_distance, closest_i, closest_j = distance, i, j
    rearranged_contour1 = np.vstack((contour1[closest_i:], contour1[:closest_i]))
    rearranged_contour2 = np.vstack((contour2[closest_j:], contour2[:closest_j]))
    return np.vstack((rearranged_contour1, rearranged_contour2)), min_distance

if __name__ == "__main__":
    check_piece_image = cv2.imread("D:/wzj/merge-contours/check_piece_image.jpg")
    cv2.imshow("check_piece_image", check_piece_image)

    check_piece_gray_image = cv2.cvtColor(check_piece_image, cv2.COLOR_BGR2GRAY)
    cv2.imshow("check_piece_gray_image", check_piece_gray_image)

    check_piece_bitwise_not_image = cv2.bitwise_not(check_piece_gray_image)
    cv2.imshow("check_piece_bitwise_not_image", check_piece_bitwise_not_image)

    # _, check_piece_thresh_image = cv2.threshold(check_piece_gray_image, 100, 255, cv2.THRESH_OTSU)
    _, check_piece_thresh_image = cv2.threshold(check_piece_bitwise_not_image, 100, 255, cv2.THRESH_OTSU)
    cv2.imshow("check_piece_thresh_image", check_piece_thresh_image)

    check_piece_contours, _ = cv2.findContours(check_piece_thresh_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    check_piece_contours_image = check_piece_image.copy()
    cv2.drawContours(check_piece_contours_image, check_piece_contours, -1, (0, 255, 0), 2)
    cv2.imshow("check_piece_contours_image", check_piece_contours_image)

    new_merged_check_piece_contours = merge_contours(check_piece_contours)
    check_piece_contours_image_by_merged = check_piece_image.copy()
    cv2.drawContours(check_piece_contours_image_by_merged, new_merged_check_piece_contours, -1, (0, 255, 0), 2)
    cv2.imshow("check_piece_contours_image_by_merged", check_piece_contours_image_by_merged)

    cv2.waitKey(0)

example_result

题外话

轮廓的合并，我有尝试其他，例如最初直接堆叠，发现效果并不好，而且无法控制轮廓的顺序；后面又考虑近似轮廓以及凸包，但关于它的一些扩展并不理想就是了；如果你们对轮廓合并有些其他的想法，可以在下面的评论中指出，欢迎指导
这里贴一些其他轮廓处理的截图

直接堆叠

all_check_piece_contours = np.vstack((check_piece_contours))
check_piece_contours_image_0 = check_piece_image.copy()
cv2.drawContours(check_piece_contours_image_0, [all_check_piece_contours], -1, (0, 255, 0), 2)
cv2.imshow("check_piece_contours_image_0", check_piece_contours_image_0)

check_piece_contours_image_0

近似轮廓

epsilon = 0.01 * cv2.arcLength(all_check_piece_contours, True)
approx_contour = cv2.approxPolyDP(all_check_piece_contours, epsilon, True)
check_piece_contours_image_1 = check_piece_image.copy()
cv2.drawContours(check_piece_contours_image_1, [approx_contour], -1, (0, 255, 0), 2)
cv2.imshow("check_piece_contours_image_1", check_piece_contours_image_1)

check_piece_contours_image_1

凸包处理的轮廓

hull = cv2.convexHull(all_check_piece_contours)
check_piece_contours_image_2 = check_piece_image.copy()
cv2.drawContours(check_piece_contours_image_2, [hull], -1, (0, 255, 0), 2)
cv2.imshow("check_piece_contours_image_2", check_piece_contours_image_2)