python用传统代码对图像进行处理识别，可以和神经网络代码做对比

用神经网络对图像进行识别时，可以先用传统算法对所得到的图像进行对比，用处理结果和后续的神经网络识别结果进行对比处理。

对图像二值化处理代码

import cv2 as cv
import numpy as np

def watershed_demo(image):
    # remove noise if any
    print(src.shape)#图像维度(348, 500, 3)
    blur = cv.pyrMeanShiftFiltering(image,10,100)#均值偏移滤波去噪
    gray = cv.cvtColor(blur,cv.COLOR_BGR2GRAY)  #获取灰度图像

    ret,binary = cv.threshold(gray,0,255,cv.THRESH_BINARY|cv.THRESH_OTSU)
    cv.imshow("binary", binary)
    #形态学操作，进一步消除图像中噪点
    kernel = cv.getStructuringElement(cv.MORPH_RECT,(3,3))#卷积核
    mb = cv.morphologyEx(binary,cv.MORPH_OPEN,kernel,iterations=2)  #iterations连续两次开操作
    sure_bg = cv.dilate(mb,kernel,iterations=3) #3次膨胀,可以获取到大部分都是背景的区域
    cv.imshow("sure_bg",sure_bg)
    #距离变换
    dist = cv.distanceTransform(mb,cv.DIST_L2,5)
    cv.imshow("dist",dist)
    dist_output = cv.normalize(dist,0,1.0,cv.NORM_MINMAX)
    # print(mb[150][120:140])
    # print(dist[150][120:140])
    # print(dist_output[150][120:140])
    cv.imshow("distinct-t",dist_output*50)
    ret, sure_fg = cv.threshold(dist,dist.max()*0.6,255,cv.THRESH_BINARY)
    cv.imshow("sure_fg",sure_fg)
    # print(sure_fg[150][120:140])
    # print(sure_bg[150][120:140])
    #获取未知区域
    surface_fg = np.uint8(sure_fg)  #保持色彩空间一致才能进行运算，现在是背景空间为整型空间，前景为浮点型空间，所以进行转换
    unknown = cv.subtract(sure_bg,surface_fg)#种子区域除外的区域=膨胀结果-种子
    cv.imshow("unkown",unknown)
    #获取maskers,在markers中含有种子区域
    ret,markers = cv.connectedComponents(surface_fg)
    #print(ret)

    #分水岭变换
    markers = markers + 1
    markers[unknown==255] = 0

    markers = cv.watershed(image,markers=markers)
    image[markers==-1] = [0,0,255]

    #cv.imshow("result",image)

src = cv.imread("342.jpg")  #读取图片位置
cv.namedWindow("input image", cv.WINDOW_AUTOSIZE)
cv.imshow("input image",src)    #通过名字将图像和窗口联系

watershed_demo(src)

cv.waitKey(0)   #等待用户操作，里面等待参数是毫秒，我们填写0，代表是永远，等待用户操作
cv.destroyAllWindows()  #销毁所有窗口

对图像进行分水岭处理

import numpy as np
import cv2

# 读取图像并转换为灰度图像
img = cv2.imread('667.jpg')
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)


# 使用sobel算子计算图像梯度
grad_x = cv2.Sobel(gray,cv2.CV_32F,1,0,ksize=3)
grad_y = cv2.Sobel(gray,cv2.CV_32F,0,1,ksize=3)
grad = cv2.subtract(grad_x,grad_y)
grad = cv2.convertScaleAbs(grad)

# 执行二值化处理
_,binary = cv2.threshold(grad,0,255,cv2.THRESH_BINARY|cv2.THRESH_OTSU)

# 对二值化图像执行闭运算
kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(3,3))
closed = cv2.morphologyEx(binary,cv2.MORPH_CLOSE,kernel)

# 执行开运算以消除所有小的白色区域
opening = cv2.morphologyEx(closed,cv2.MORPH_OPEN,kernel)

# 距离变换
dist_transform = cv2.distanceTransform(opening,cv2.DIST_L2,5)
_, sure_fg = cv2.threshold(dist_transform, 0.7*dist_transform.max(), 255, 0)

# 背景重构
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(opening, sure_fg)
_, markers = cv2.connectedComponents(sure_fg)
markers = markers + 1
markers[unknown==255] = 0
markers = cv2.watershed(img, markers)

# 显示分割结果
img[markers == -1] = [255,0,0]

cv2.imshow('Input Image', img)
cv2.imshow('Binary Image', binary)
cv2.imshow('Closed Image', closed)
cv2.imshow('Opening Image', opening)
cv2.imshow('Sure Foreground Image', sure_fg)
cv2.imshow('Uknown Image', unknown)
#cv2.imshow('Markers Image', markers)

cv2.waitKey(0)
cv2.destroyAllWindows()