第三方包安装

pip install pyclipper

1. 轮廓点等距离外扩

def equidistant_zoom_contour(contour, margin):
    """
    等距离缩放多边形轮廓点
    :param contour: 一个图形的轮廓格式[[[x1, x2]],...],shape是(-1, 1, 2)
    :param margin: 轮廓外扩的像素距离，margin正数是外扩，负数是缩小
    :return: 外扩后的轮廓点
    """
    pco = pyclipper.PyclipperOffset()
    ##### 参数限制，默认成2这里设置大一些，主要是用于多边形的尖角是否用圆角代替
    pco.MiterLimit = 10
    contour = contour[:, 0, :]
    pco.AddPath(contour, pyclipper.JT_MITER, pyclipper.ET_CLOSEDPOLYGON)
    solution = pco.Execute(margin)
    solution = np.array(solution).reshape(-1, 1, 2).astype(int)
    return solution

调试用例

import pyclipper
import math
from shapely.geometry import LineString, Polygon, MultiLineString, Point, MultiPoint

poly = np.array([[[200, 200]], [[200, 300]], [[400, 350]], [[350, 200]], [[300, 200]], [[200, 100]]])
contour1 = equidistant_zoom_contour(poly, 20)
img = np.zeros((500, 500, 3))
cv2.polylines(img, [poly], True, (0, 0, 255), 3)
cv2.polylines(img, [contour1], True, (0, 255, 0), 3)

结果展示：

2. 轮廓点等比例缩放

def perimeter(poly):
    p = 0
    nums = poly.shape[0]
    for i in range(nums):
        p += abs(np.linalg.norm(poly[i % nums] - poly[(i + 1) % nums]))
    return p

def proportional_zoom_contour(contour, ratio):
    """
    多边形轮廓点按照比例进行缩放
    :param contour: 一个图形的轮廓格式[[[x1, x2]],...],shape是(-1, 1, 2)
    :param ratio: 缩放的比例，如果大于1是放大小于1是缩小
    :return:
    """
    poly = contour[:, 0, :]
    area_poly = abs(pyclipper.Area(poly))
    perimeter_poly = perimeter(poly)
    poly_s = []
    pco = pyclipper.PyclipperOffset()
    pco.MiterLimit = 10
    if perimeter_poly:
        d = area_poly * (1 - ratio * ratio) / perimeter_poly
        pco.AddPath(poly, pyclipper.JT_MITER, pyclipper.ET_CLOSEDPOLYGON)
        poly_s = pco.Execute(-d)
    poly_s = np.array(poly_s).reshape(-1, 1, 2).astype(int)

    return poly_s

测试用例：

import pyclipper
import math
from shapely.geometry import LineString, Polygon, MultiLineString, Point, MultiPoint
poly = np.array([[[200, 200]], [[200, 300]], [[400, 350]], [[350, 200]], [[300, 200]], [[200, 100]]])
contour1 = proportional_zoom_contour(poly, 1.5)
img = np.zeros((500, 500, 3))
cv2.polylines(img, [contour1], True, (0, 255, 0), 3)
cv2.polylines(img, [poly], True, (0, 0, 255), 3)

其中， pco.MiterLimit = 10这个参数默认是2，如果是默认的值结果图第一个，改成10的话，结果图就是第二个，是一个尖角的区别

3. 图形轮廓的旋转

# 获取一个形状的质心
def get_centroid(coord):
    coord = np.array(coord)
    shape = coord.shape
    if len(shape) == 1 and len(coord) == 2:  # point
        return coord
    if len(shape) == 1 and len(coord) == 4:  # bounding box
        return tuple([(coord[0] + coord[2]) // 2, (coord[1] + coord[3]) // 2])
    elif len(shape) == 2 and shape[-1] == 2:
        if shape[0] == 2:  # 如果是直线
            cen = LineString(coord).centroid
        else:
            cen = Polygon(coord).centroid
        return tuple(map(int, [cen.x, cen.y]))
    elif len(shape) == 3 and shape[1:] == (1, 2):  # contour
        cen = Polygon(coord.squeeze()).centroid
        return tuple(map(int, [cen.x, cen.y]))
    else:
        raise Exception('coordinate error, must be bbox or contour shape:{}'.format(coord))


def point_Srotate(im_w, im_h, angle, spin_point, origin_point):
    """
    :param im_w: 原始点所在的图片的宽度
    :param im_h: 原始点所在的图片的高度
    :param angle: 旋转的角度
    :param spin_point: 旋转的点
    :param origin_point: 参考点
    :return: 旋转过后的点
    """
    row, col = im_h, im_w
    # P(x1, y1)，绕某个像素点Q(x2, y2)
    x1, y1 = spin_point
    x2, y2 = origin_point
    y1 = row - y1
    y2 = row - y2
    x = (x1 - x2) * math.cos(math.pi / 180.0 * angle) - (y1 - y2) * math.sin(math.pi / 180.0 * angle) + x2
    y = (x1 - x2) * math.sin(math.pi / 180.0 * angle) + (y1 - y2) * math.cos(math.pi / 180.0 * angle) + y2
    x = x
    y = row - y

    return [x, y]

调用示例：

import pyclipper
import math
from shapely.geometry import LineString, Polygon, MultiLineString, Point, MultiPoint
# 以多边形轮廓的质心为参照点进行旋转
poly = np.array([[[200, 200]], [[200, 300]], [[400, 350]], [[350, 200]], [[300, 200]], [[200, 100]]])

origin_point = get_centroid(poly)
spin_list = []
for con in poly:
    print('con', con)
    new = point_Srotate(500, 500, 50, con[0], origin_point)
    spin_list.append(new)
spin_con = np.array(spin_list).reshape(-1, 1, 2).astype(int)
img = np.zeros((500, 500, 3))
cv2.polylines(img, [spin_con], True, (0, 255, 0), 3)
cv2.polylines(img, [poly], True, (0, 0, 255), 3)

结果：

4. 其他外扩函数

def extend_contour2(contour, margin):
    # 每个点相对于质心进行外扩一定的距离
    """
    :param contour: 轮廓点集合
    :param margin: 外扩的距离
    :return: 外扩后的轮廓点集
    """
    #### 求该轮廓的质心 ####
    gravity_point = get_centroid(contour)
    #### 获取最左下的点 ####
    # min_x = np.minimum(contour)
    #### 计算所有的轮廓点与质心所组成的向量，计算向量的模
    vector_arr = contour - np.array(gravity_point)
    vector_length = np.linalg.norm(vector_arr, axis=2)
    #### 计算所有的点针对对外扩的像素需要放大多少倍
    ratio = 1 + margin / vector_length
    ratio = np.concatenate([ratio, ratio], axis=1)
    #### 进行坐标的缩放
    contour_ext = (vector_arr[:, 0, :] * ratio + np.array(gravity_point)).reshape(-1, 1, 2)
    contour_ext = contour_ext.astype(int)
    return contour_ext

def coordinate_conversion(reference_point, contour, ratio):
    # 对凸多边形有用，对凹多边形容易变形，成比例缩放轮廓
    """
    :param reference_point: 参照点的坐标
    :param contour: 图像的轮廓点
    :param ratio: 缩放的比例
    :return: 以参照点不变将轮廓点获取缩放后的轮廓点坐标
    """
    contour_trans_array = (contour - np.array(reference_point)) * ratio + np.array(reference_point)
    contour_trans_array = contour_trans_array.astype(int)
    return contour_trans_array