Susan角点检测python实现（边缘检测、角点检测、重心计算、非极大值抑制）

#coding=utf8 import numpy as np from PIL import Image import matplotlib.pyplot as plt import matplotlib.cm as cm import math import cv2 import os,sys import scipy.ndimage import time import scipy def get_susan_mask(): mask=np.ones((7,7)) mask[0,0]=0;mask[0,1]=0;mask[0,5]=0;mask[0,6]=0; mask[1,0]=0;mask[1,6]=0; mask[5,0]=0;mask[5,6]=0;mask[6,0]=0;mask[6,1]=0; mask[6,5]=0;mask[6,6]=0; return mask def susan_corner_detect(img_src,t=10): susan_mask = get_susan_mask() img = img_src.copy() row_s,col_s = 3,3 row_e,col_e = img_src.shape[0]-3,img.shape[1]-3 n_max = 0 n_arr=37*np.ones(img.shape) # 初始认为没有角点 for r in range(row_s,row_e):#遍历所有行 for c in range(col_s,col_e):#遍历所有列 susan_zone = img[r-3:r+3+1,c-3:c+3+1]#获取矩形区域 susan_zone = susan_zone[susan_mask!=0]#使用mask截取圆形区域 r0 = img[r,c] similarity = np.exp(-((1.0*susan_zone-r0)/t)**6 ) n=np.sum(similarity) if n>n_max: n_max = n n_arr[r,c] = n g = n_max /2 R = np.zeros(img.shape) index = n_arr<g #小于g，认为是可能的角点，越小，可能性越大 R[index] = g-n_arr[index] # 取反，所以R越大，是角点的可能性越大 plt.figure() plt.title("edge") plt.imshow((6.37 * n_arr).astype(np.uint8), cmap=cm.gray) return R def gravity_filter(img_src,corner_src,t=10,F=1.5): x_label = np.zeros((7,7)) y_label = np.zeros((7,7)) x_label[:,0]=-3;x_label[:,1]=-2;x_label[:,2]=-1; x_label[:, -1] = 3; x_label[:, -2] = 2; x_label[:, -3] = 1; y_label[0,:]=-3;y_label[1,:]=-2;y_label[2,:]=-1; y_label[4, :] = 1;y_label[5, :] = 2; y_label[6, :] = 3; print(x_label,"\r\n",y_label) #查看矩形区域内x、y轴信息 img = img_src.copy() row_s, col_s = 3, 3 row_e, col_e = img_src.shape[0] - 3, img.shape[1] - 3 corner = corner_src.copy() susan_mask = get_susan_mask() for r in range(row_s,row_e): for c in range(col_s,col_e): if corner[r,c] ==0:#对于不是角点的位置，就没必要进行后续计算了 continue susan_zone = img[r-3:r+3+1,c-3:c+3+1]#获取矩形区域 r0 = img[r,c] similarity = np.exp(-((1.0*susan_zone-r0)/t)**6 ) g_x = np.sum(similarity[susan_mask==1]*x_label[susan_mask==1] )/np.sum(similarity[susan_mask==1])#使用mask截取圆形区域 g_y = np.sum(similarity[susan_mask == 1] * y_label[susan_mask == 1]) / np.sum(similarity[susan_mask == 1])#使用mask截取圆形区域 distance = np.sqrt(g_x**2+g_y**2) if distance<F: corner[r,c] = 0 return corner def corner_nms(corner,kernal=3): out = corner.copy() row_s = int(kernal/2) row_e = out.shape[0] - int(kernal/2) col_s,col_e = int(kernal/2),out.shape[1] - int(kernal/2) for r in range(row_s,row_e): for c in range(col_s,col_e): if corner[r,c]==0: #不是可能的角点 continue zone = corner[r-int(kernal/2):r+int(kernal/2)+1,c-int(kernal/2):c+int(kernal/2)+1] index = corner[r,c]<zone (x,y) = np.where(index==True) if len(x)>0 : #说明corner[r,c]不是最大，直接归零将其抑制 out[r,c] = 0 else: out[r,c] = 255 return out if __name__ == '__main__': img_src = cv2.imread('susan_input1.png',-1) if len(img_src.shape)==3: img_src = cv2.cvtColor(img_src,cv2.COLOR_BGR2GRAY) corner = susan_corner_detect(img_src) img_show = cv2.cvtColor(img_src, cv2.COLOR_GRAY2BGR) img_show[corner != 0] = (255, 0, 0) plt.figure() plt.title("original corners") plt.imshow(img_show, cmap=cm.gray) cor_g = gravity_filter(img_src, corner) img_show2 = cv2.cvtColor(img_src, cv2.COLOR_GRAY2BGR) img_show2[cor_g != 0] = (255, 0, 0) plt.figure() plt.title("corners-gravity ") plt.imshow(img_show2, cmap=cm.gray) cor_g_nms = corner_nms(cor_g) img_show3 = cv2.cvtColor(img_src, cv2.COLOR_GRAY2BGR) img_show3[cor_g_nms != 0] = (255, 0, 0) plt.figure() plt.title("corners-gravity-nms ") plt.imshow(img_show3, cmap=cm.gray) plt.show() print('emd')