激光条纹中心提取——Zhang-Suen法python

Zhang-Suen法

原理—— Zhang-Suen法
代码——python代码

原理—— Zhang-Suen法

细化法（又称形态学骨架法）是通过对光条纹不断地进行腐蚀操作，剥离光条纹边界，得到单像素宽度的光条纹连通线（又称骨架）的形态学处理方法。该方法是通过细化技术，将光条纹区域的细化曲线作为光条纹中心线。由于细化法是基于形态学的方法，只是对光条纹骨架进行提取，没有考虑到光条纹的横截面灰度分布特点。因此，细化法提取的光条纹中心线精度有限。另一方面，由于该方法需要大量时间来进行反复的细化操作，提取算法的运算速度被大大降低。
细化的方法有很多种，下面介绍一种常用的Zhang-Suen细化算法。细化算法的一般分为以下四个步骤，
（1）标记将被删除的边界点；
（2）删除做了标记的点；
（3）继续标记将被删除的剩余的边界点；
（4）删除标记过的点。反复应用这个基本过程，直到再也没有被删除的点为止，此时算法终止，生成了该区域的骨架。
假设目标像素点标记为1，背景像素点标记为0，在图上图中的八个邻域如图所示，Zhang-Suen算法第一步循环所有前景像素点，对满足如下公式（1-5）的像素标记点进行删除。第二步和第一步类似，满足公式（1-6）的像素标记点记为删除。循环上述两个步骤，直到两步中没有像素标记点记为删除为止，输出图像即为二值图像细化后的骨架。

其中 N(P0)表示八个邻域中非零像素标记点的个数（二值图像只有0和1）， S(P0)表示八个邻域中，按照顺时针方向，相邻两个像素出现0→1的次数，。
细化法由于需要做多次膨胀腐蚀操作，耗时较多，处理速度较慢，另外细化法对二值化的要求极高，往往因为噪声的干扰，将图像二值化会出现较大的误差。细化法属于形态学范畴，没有利用到灰度值之间的信息，因此对于复杂条件下的激光曲线提取效果较差。

代码——python代码

# 作用： zhang-sufa
# 作者： yeluo
import cv2
import numpy as np
import time
import matplotlib.pyplot as pltdef ROI(img):indexXY = np.argwhere(img > 0)minxy = np.min(indexXY, axis=0)maxxy = np.max(indexXY, axis=0)return minxy,maxxy
def neighbours(x,y,img):i = imgx1,y1,x_1, y_1 = x+1, y-1, x-1, y+1return [i[y1][x],  i[y1][x1],   i[y][x1],  i[y_1][x1],  # P2,P3,P4,P5i[y_1][x], i[y_1][x_1], i[y][x_1], i[y1][x_1]]  # P6,P7,P8,P9
def transitions(neighbours):n = neighbours + neighbours[0:1]  # P2, ... P9, P2return sum((n1, n2) == (0, 1) for n1, n2 in zip(n, n[1:]))
def ZhangSuenPlus(image):"""运行时间55s:param image::return:"""changing1 = changing2 = [(-1, -1)]while changing1 or changing2:# Step 1changing1 = []for y in range(1, len(image) - 1):for x in range(1, len(image[0]) - 1):P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, image)if (image[y][x] == 1 and    # (Condition 0)P4 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P6 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing1.append((x,y))for x, y in changing1: image[y][x] = 0# Step 2changing2 = []for y in range(1, len(image) - 1):for x in range(1, len(image[0]) - 1):P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, image)if (image[y][x] == 1 and    # (Condition 0)P2 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P8 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing2.append((x,y))for x, y in changing2: image[y][x] = 0#print changing1#print changing2flags = image>0image[flags] = 255#cv2.imshow("res",image)return image
def ZhangSuenPlus02(image):"""# 运行时间12.135秒:param image::return:"""indexXY = np.argwhere(image>0)minxy = np.min(indexXY,axis=0)maxxy = np.max(indexXY,axis=0)roi = image[minxy[0]-1:maxxy[0]+2,minxy[1]-1:maxxy[1]+2]changing1 = changing2 = [(-1, -1)]while changing1 or changing2:# Step 1changing1 = []for y in range(1, len(roi) - 1):for x in range(1, len(roi[0]) - 1):if roi[y][x] == 1:P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, roi)if (    # (Condition 0)P4 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P6 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing1.append((x,y))for x, y in changing1: roi[y][x] = 0# Step 2changing2 = []for y in range(1, len(roi) - 1):for x in range(1, len(roi[0]) - 1):if roi[y][x] == 1:P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, roi)if (   # (Condition 0)P2 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P8 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing2.append((x,y))for x, y in changing2: roi[y][x] = 0#print changing1#print changing2flags = roi>0roi[flags] = 255#cv2.imshow("res",image)return image
def ZhangSuenPlus03(image):"""# 运行时间9秒:param image::return:"""indexXY = np.argwhere(image>0)minxy = np.min(indexXY,axis=0)maxxy = np.max(indexXY,axis=0)roi = image[minxy[0]-1:maxxy[0]+2,minxy[1]-1:maxxy[1]+2]changing1 = changing2 = [(-1, -1)]while changing1 or changing2:indexXY = np.argwhere(roi>0)minxy = np.min(indexXY, axis=0)maxxy = np.max(indexXY, axis=0)roi = roi[minxy[0] - 1:maxxy[0] + 2, minxy[1] - 1:maxxy[1] + 2]# Step 1changing1 = []for y in range(1, len(roi) - 1):for x in range(1, len(roi[0]) - 1):if roi[y][x] == 1:P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, roi)if (    # (Condition 0)P4 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P6 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing1.append((x,y))for x, y in changing1: roi[y][x] = 0# Step 2changing2 = []for y in range(1, len(roi) - 1):for x in range(1, len(roi[0]) - 1):if roi[y][x] == 1:P2,P3,P4,P5,P6,P7,P8,P9 = n = neighbours(x, y, roi)if (   # (Condition 0)P2 * P6 * P8 == 0 and   # Condition 4P2 * P4 * P8 == 0 and   # Condition 3transitions(n) == 1 and # Condition 22 <= sum(n) <= 6):      # Condition 1changing2.append((x,y))for x, y in changing2: roi[y][x] = 0#print changing1#print changing2flags = roi>0roi[flags] = 255#cv2.imshow("res",image)return imagedef ZhangSuen_Bad(img):"""将灰度值转化为0和1，企图加速计算过程运行时间21.28sthresh计算失败:param img::return:"""copyMat = img.copy()k = 0row,col= img.shaperow = row-1col = col-1while(True):k= k+1stop= False# step1for i in range(1,row):for j in range(1,col):if img[i,j]>0:print(">0")p1 = 1 if img[i,j]>0 else 0p2 = 1 if img[i-1,j]>0 else 0p3 = 1 if img[i-1,j+1]>0 else 0p4 = 1 if img[i, j+1] > 0 else 0p5 = 1 if img[i+1, j+1] > 0 else 0p6 = 1 if img[i+1, j] > 0 else 0p7 = 1 if img[i+1, j-1] > 0 else 0p8 = 1 if img[i,j-1] > 0 else 0p9 = 1 if img[i-1, j-1] > 0 else 0np1 = p2+p3+p4+p5+p6+p7+p8+p9sp2 = 1 if (p2 == 0 and p3 == 1) else 0sp3 = 1 if (p3 == 0 and p4 == 1) else 0sp4 = 1 if (p4 == 0 and p5 == 1) else 0sp5 = 1 if (p5 == 0 and p6 == 1) else 0sp6 = 1 if (p6 == 0 and p7 == 1) else 0sp7 = 1 if (p7 == 0 and p8 == 1) else 0sp8 = 1 if (p8 == 0 and p9 == 1) else 0sp9 = 1 if (p9 == 0 and p2 == 1) else 0sp1 = sp2 + sp3 + sp4 + sp5 + sp6 + sp7 + sp8 + sp9if np1>=2 and np1<=6 and sp1==1 and(p2*p4*p6)==0 and (p4*p6*p8)==0:stop = TruecopyMat[i,j] = 0print("success")img = copyMat.copy()# step2for i in range(1,row):for j in range(1,col):if img[i,j]>0:print(">>")p2 = 1 if img[i - 1, j] > 0 else 0p3 = 1 if img[i - 1, j + 1] > 0 else 0p4 = 1 if img[i, j + 1] > 0 else 0p5 = 1 if img[i + 1, j + 1] > 0 else 0p6 = 1 if img[i + 1, j] > 0 else 0p7 = 1 if img[i + 1, j - 1] > 0 else 0p8 = 1 if img[i, j - 1] > 0 else 0p9 = 1 if img[i - 1, j - 1] > 0 else 0np1 = p2 + p3 + p4 + p5 + p6 + p7 + p8 + p9sp2 = 1 if (p2 == 0 and p3 == 1) else 0sp3 = 1 if (p3 == 0 and p4 == 1) else 0sp4 = 1 if (p4 == 0 and p5 == 1) else 0sp5 = 1 if (p5 == 0 and p6 == 1) else 0sp6 = 1 if (p6 == 0 and p7 == 1) else 0sp7 = 1 if (p7 == 0 and p8 == 1) else 0sp8 = 1 if (p8 == 0 and p9 == 1) else 0sp9 = 1 if (p9 == 0 and p2 == 1) else 0sp1 = sp2 + sp3 + sp4 + sp5 + sp6 + sp7 + sp8 + sp9if np1 >= 2 and np1 <= 6 and sp1 == 1 and (p2*p4*p8) == 0 and (p2*p6*p8) == 0:stop = TruecopyMat[i,j] = 0print("success")img = copyMat.copy()if(not stop):breakresImg = copyMat.copy()flags = resImg>0resImg[flags] = 255#print(k)# cv2.imshow("res",resImg)return resImgdef ZhangSuen(img):"""运行时间20.7s:param img::return:"""#indexXY = np.argwhere(img>0)#minxy = np.min(indexXY,axis=0)#maxxy = np.max(indexXY,axis=0)#roi = img[minxy[0]-3:maxxy[0]+4,minxy[1]-3:maxxy[1]+4]#flags = roi>0#roi[flags] = 255#cv2.imshow("roi",roi)#print(roi.shape)roi = imgk = 0row,col= roi.shapechanging1 = changing2 = [(-1, -1)]while changing1 or changing2:changing1 = []for i in range(1,row-1):for j in range(1,col-1):if roi[i,j]==1:p2 = roi[i-1,j]p3 = roi[i-1,j+1]p4 = roi[i, j+1]p5 = roi[i+1, j+1]p6 = roi[i+1, j]p7 = roi[i+1, j-1]p8 = roi[i,j-1]p9 = roi[i-1, j-1]np1 = p2+p3+p4+p5+p6+p7+p8+p9sp2 = 1 if (p2,p3)==(0,1) else 0sp3 = 1 if (p3,p4)==(0,1) else 0sp4 = 1 if (p4,p5)==(0,1) else 0sp5 = 1 if (p5,p6)==(0,1) else 0sp6 = 1 if (p6,p7)==(0,1) else 0sp7 = 1 if (p7,p8)==(0,1) else 0sp8 = 1 if (p8,p9)==(0,1) else 0sp9 = 1 if (p9,p2)==(0,1) else 0sp1 = sp2 + sp3 + sp4 + sp5 + sp6 + sp7 + sp8 + sp9if 2<=np1<=6 and sp1==1 and(p2*p4*p6)==0 and (p4*p6*p8)==0:changing1.append([i,j])for x,y in changing1:roi[x,y] = 0# step2changing2 = []for i in range(1,row-1):for j in range(1,col-1):if roi[i,j]==1:p2 = roi[i - 1, j]p3 = roi[i - 1, j + 1]p4 = roi[i, j + 1]p5 = roi[i + 1, j + 1]p6 = roi[i + 1, j]p7 = roi[i + 1, j - 1]p8 = roi[i, j - 1]p9 = roi[i - 1, j - 1]np1 = p2 + p3 + p4 + p5 + p6 + p7 + p8 + p9sp2 = 1 if (p2, p3) == (0, 1) else 0sp3 = 1 if (p3, p4) == (0, 1) else 0sp4 = 1 if (p4, p5) == (0, 1) else 0sp5 = 1 if (p5, p6) == (0, 1) else 0sp6 = 1 if (p6, p7) == (0, 1) else 0sp7 = 1 if (p7, p8) == (0, 1) else 0sp8 = 1 if (p8, p9) == (0, 1) else 0sp9 = 1 if (p9, p2) == (0, 1) else 0sp1 = sp2 + sp3 + sp4 + sp5 + sp6 + sp7 + sp8 + sp9if 2<=np1<= 6 and sp1 == 1 and (p2*p4*p8) == 0 and (p2*p6*p8) == 0:#roi[i,j] = 0changing2.append([i,j])#print("success")for x,y in changing2:roi[x,y] = 0flags = roi>0roi[flags] = 255return roidef Draw():plt.figure()plt.subplot(131)plt.imshow(cv2.cvtColor(img,cv2.COLOR_BGR2RGB))plt.title("origin image")plt.axis("off")plt.subplot(132)# plt.imshow(res,"gray")plt.title("res image")plt.axis("off")plt.subplot(133)# plt.imshow(resP,"gray")plt.title("resP image")plt.axis("off")plt.show()
# Draw()def XiHua(img):kernel_d = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))image_d = cv2.dilate(image, kernel_d, iterations=8)# cv2.namedWindow('dilate', cv2.WINDOW_NORMAL)# cv2.imshow('dilate', image_d)kernel_e = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))image_e = cv2.erode(image_d, kernel_e)image_ske = morphology.skeletonize(image_e)image_e = np.multiply(image_e, image_ske)return image_eif __name__ == "__main__":import osimport timefrom skimage import morphologyimage_path = './image/'save_path = './paper/zhang-suen/'sum_time = 0if not os.path.isdir(save_path): os.makedirs(save_path)for img in os.listdir(image_path):image = cv2.imread(os.path.join(image_path, img))gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)ret, thresh = cv2.threshold(gray, 100, 255, cv2.THRESH_BINARY)time1 = time.time()result = ZhangSuen(thresh)time2 = time.time()temp_time = time2 - time1sum_time += temp_time# print("one image time: ", temp_time)# cv2.namedWindow('result', 0)# cv2.imshow('result', result)# cv2.waitKey(10)cv2.imwrite(os.path.join(save_path, img), result)per_time = sum_time / len(os.listdir(image_path))print("average image time:", per_time)