边缘检测中的非极大值抑制算法的python实现

非极大值抑制算法实现，参考这位博主的：
https://blog.csdn.net/qq_45717425/article/details/120641010

import cv2
import numpy as np
import matplotlib.pyplot as plt
def NMS(grad,gradx,grady):
    [h,w]=np.shape(grad)
    weight=np.zeros((h,w))
    result=np.zeros((h,w))
    grad1 =np.zeros((h,w))
    grad2 =np.zeros((h,w))
    grad3 =np.zeros((h,w))
    grad4 =np.zeros((h,w))
    gradTemp1 =np.zeros((h,w))
    gradTemp2 =np.zeros((h,w))
    #留出一圈像素进行缓冲
    for i in range(2,h-1):
        for j in range(2,w-1):
            if grad[i,j]==0:
                result[i,j]=0
            else:
                if abs(grady[i,j]>gradx[i,j]):
                    weight[i,j]=gradx[i,j]/grady[i,j]
                    grad2[i,j]=grad[i-1,j]
                    grad4[i,j]=grad[i+1,j]
                    if grady[i,j]*gradx[i,j]>0:
                        grad1[i,j]=grad[i-1,j-1]
                        grad3[i,j]=grad[i+1,j+1]
                    else:
                        grad1[i,j]=grad[i-1,j+1]
                        grad3[i,j]=grad[i+1,j-1]
                else:
                    weight[i,j]=grady[i,j]/gradx[i,j]
                    grad2[i,j]=grad[i,j-1]
                    grad4[i,j]=grad[i,j+1]                    
                    if grady[i,j]*gradx[i,j]>0:
                        grad1[i,j]=grad[i-1,j-1]
                        grad3[i,j]=grad[i+1,j+1]
                    else:
                        grad1[i,j]=grad[i+1,j-1]
                        grad3[i,j]=grad[i-1,j+1]
                gradTemp1[i,j]= weight[i,j]*grad1[i,j]+(1-weight[i,j])*grad2[i,j]
                gradTemp2[i,j]=weight[i,j]*grad3[i,j]+(1-weight[i,j])*grad4[i,j]
                if grad[i,j] >= gradTemp1[i,j] and grad[i,j] >= gradTemp2[i,j]:
                    result[i,j] = grad[i,j]
                else:
                    result[i,j] =0
    return result

附上手动sobel代码

#纯手搓sobel
def sobel_filter2(img):
    sp = img.shape
    height = sp[0]
    weight = sp[1]
    sx = np.array([[-1,0,1],[-2,0,2],[-1,0,1]])
    sy = np.array([[-1,-2,-1],[0,0,0],[1,2,1]])
    dSobel = np.zeros((height,weight))
    Gx = np.zeros(img.shape)
    Gy = np.zeros(img.shape)
    for i in range(height-2):
        for j in range(weight-2):
            Gx[i + 1, j + 1] = abs(np.sum(img[i:i + 3, j:j + 3] * sx))
            Gy[i + 1, j + 1] = abs(np.sum(img[i:i + 3, j:j + 3] * sy))
            dSobel[i+1, j+1] =np.sqrt(Gx[i+1, j+1]**2+Gy[i+1, j+1]**2)
    return dSobel,Gx,Gy 

调用函数运行

img = cv2.imread(r"C:\Users\Cosine\Desktop\175.jpg", cv2.IMREAD_GRAYSCALE)
grad,gradx,grady= sobel_filter2(img)
NMSfigure=NMS(grad,gradx,grady)
sobel=np.array(grad,dtype=np.uint8)
NMSfigure=np.array(NMSfigure,dtype=np.uint8)
cv2.imshow('Original Image', img)
cv2.imshow('sobel', sobel)
cv2.imshow('NMSfigure', NMSfigure)
cv2.waitKey(0)
cv2.destroyAllWindows()

结果