python cv2.HoughCircles 霍夫圆检测

cv2提供了一种圆检测的方法：HoughCircles。该函数的返回结果与参数设置有很大的关系。
检测的图像时9枚钱币，分别使用了阈值（大津法和三角法）、均值偏移滤波以及未处理图像。实验的结果是只要调整param1和param2两个参数，上述方法都能准确的识别图像中的圆形。与圆最贴切的是大津法阈值。使用该方法同时需要使用cv2.THRESHOLD_TRUNC。

1. HoughCircles说明

函数定义如下：

HoughCircles(image, method, dp, minDist, circles=None, param1=None, param2=None, minRadius=None, maxRadius=None)

参数	含义
image	原始图像
method	目前只支持cv2.HOUGH_GRADIENT
dp	图像解析的反向比例。1为原始大小，2为原始大小的一半
minDist	圆心之间的最小距离。过小会增加圆的误判，过大会丢失存在的圆
param1	Canny检测器的高阈值
param2	检测阶段圆心的累加器阈值。越小的话，会增加不存在的圆；越大的话，则检测到的圆就更加接近完美的圆形
minRadius	检测的最小圆的半径
maxRadius	检测的最大圆的半径

2. 代码

# coding:utf8

import cv2
import numpy as np


def row_method(src):
    image = np.array(src)
    cimage = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)  # 灰度图
    circles = cv2.HoughCircles(cimage, cv2.HOUGH_GRADIENT, 1, 40, param1=250, param2=58, minRadius=0)
    circles = np.uint16(np.around(circles))  # 取整
    for i in circles[0, :]:
        cv2.circle(image, (i[0], i[1]), i[2], (0, 0, 255), 2)  # 在原图上画圆，圆心，半径，颜色，线框
        cv2.circle(image, (i[0], i[1]), 2, (255, 0, 0), 2)  # 画圆心
    cv2.putText(image, "param1=250, param2=58", (20, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    cv2.imshow("row_circles", image)


def threshold_OTSU_method(src):
    image = np.array(src)
    cimage = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)  # 灰度图
    th, dst = cv2.threshold(cimage, 200, 255, cv2.THRESH_BINARY + cv2.THRESH_TRUNC + cv2.THRESH_OTSU)
    circles = cv2.HoughCircles(dst, cv2.HOUGH_GRADIENT, 1, 40, param1=50, param2=47, minRadius=0)
    circles = np.uint16(np.around(circles))  # 取整
    for i in circles[0, :]:
        cv2.circle(image, (i[0], i[1]), i[2], (0, 0, 255), 2)  # 在原图上画圆，圆心，半径，颜色，线框
        cv2.circle(image, (i[0], i[1]), 2, (255, 0, 0), 2)  # 画圆心
    cv2.putText(image, "param1=50, param2=47", (20, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    cv2.imshow("otsu_circles", image)


def threshold_triangle_method(src):
    image = np.array(src)
    cimage = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)  # 灰度图
    th, dst = cv2.threshold(cimage, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_TRIANGLE)
    circles = cv2.HoughCircles(dst, cv2.HOUGH_GRADIENT, 1, 40, param1=50, param2=17, minRadius=0)
    circles = np.uint16(np.around(circles))  # 取整
    for i in circles[0, :]:
        cv2.circle(image, (i[0], i[1]), i[2], (0, 0, 255), 2)  # 在原图上画圆，圆心，半径，颜色，线框
        cv2.circle(image, (i[0], i[1]), 2, (255, 0, 0), 2)  # 画圆心
    cv2.putText(image, "param1=50, param2=17", (20, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    cv2.imshow("triangle_circles", image)


def mean_circles(src):
    image = np.array(src)
    dst = cv2.pyrMeanShiftFiltering(image, 10, 100)  # 均值偏移滤波
    cimage = cv2.cvtColor(dst, cv2.COLOR_BGR2GRAY)  # 灰度图
    circles = cv2.HoughCircles(cimage, cv2.HOUGH_GRADIENT, 1, 40, param1=50, param2=20, minRadius=0)
    circles = np.uint16(np.around(circles))  # 取整
    for i in circles[0, :]:
        cv2.circle(image, (i[0], i[1]), i[2], (0, 0, 255), 2)  # 在原图上画圆，圆心，半径，颜色，线框
        cv2.circle(image, (i[0], i[1]), 2, (255, 0, 0), 2)  # 画圆心

    cv2.putText(image, "param1=50, param2=20", (20, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    cv2.imshow("mean_circles", image)


src = cv2.imread("circle.png")  # 读取图片位置
cv2.namedWindow("input image", cv2.WINDOW_AUTOSIZE)
cv2.imshow("input image", src)
threshold_OTSU_method(src)
threshold_triangle_method(src)
mean_circles(src)
row_method(src)
cv2.waitKey(0)
cv2.destroyAllWindows()

3.结果