基于人像的抖动进行边缘分割

具体代码如下

# -*- coding: utf-8 -*-
"""
Created on Thu Aug 15 10:48:09 2019

@author: 01
"""

#!/usr/bin/env python

import cv2
import numpy as np
ph1 = "img/1.jpg"
ph2 = "img/2.jpg"
ph3 = "img/3.jpg"
ph4 = "img/4.jpg"
ph5 = "img/5.jpg"
ph6 = "img/6.jpg"
ph7 = "img/7.jpg"
ph8 = "img/8.jpg"
threshod= 40 

s1 = cv2.imread(ph1,0)
s2 = cv2.imread(ph2,0)
s3 = cv2.imread(ph3,0)
s4 = cv2.imread(ph4,0)
s5 = cv2.imread(ph5,0)
s6 = cv2.imread(ph6,0)
s7 = cv2.imread(ph7,0)
s8 = cv2.imread(ph8,0)
emptyimg1 = np.zeros(s1.shape,np.uint8)
emptyimg2 = np.zeros(s1.shape,np.uint8)
emptyimg3 = np.zeros(s1.shape,np.uint8)
emptyimg4 = np.zeros(s1.shape,np.uint8)
emptyimg5 = np.zeros(s1.shape,np.uint8)
emptyimg6 = np.zeros(s1.shape,np.uint8)
emptyimg7 = np.zeros(s1.shape,np.uint8)
emptyimg8 = np.zeros(s1.shape,np.uint8)
emptyimg9 = np.zeros(s1.shape,np.uint8)
def pic_sub(dest,s1,s2):
    for x in range(dest.shape[0]):
        for y in range(dest.shape[1]):
            if(s2[x,y] > s1[x,y]):
                dest[x,y] = s2[x,y] - s1[x,y]
            else:
                dest[x,y] = s1[x,y] - s2[x,y]

            if(dest[x,y] < threshod):
                dest[x,y] = 0
            else:
                dest[x,y] = 255

def pic_add(dest,s1,s2):
    for x in range(dest.shape[0]):
        for y in range(dest.shape[1]):
            if(s2[x,y] > s1[x,y]):
                dest[x,y] = s2[x,y] - s1[x,y]
            else:
                dest[x,y] = s1[x,y] - s2[x,y]

            if(dest[x,y] < threshod):
                dest[x,y] = 0
            else:
                dest[x,y] = 255

pic_sub(emptyimg1,s1,s3)
pic_sub(emptyimg2,s2,s4)
pic_sub(emptyimg3,s3,s5)
pic_sub(emptyimg4,s4,s6)

pic_add(emptyimg5,emptyimg1,emptyimg2)
pic_add(emptyimg6,emptyimg3,emptyimg4)
pic_add(emptyimg7,emptyimg5,emptyimg6)
cv2.namedWindow("s1")
cv2.namedWindow("s2")
cv2.namedWindow("result")

cv2.imshow("s1",emptyimg1)
cv2.imshow("s2",emptyimg2)
cv2.imshow("result",emptyimg7)

cv2.waitKey(0)
cv2.destroyAllWindows()

如果有一个视频，可以从中拿出相邻的两帧图像，最简单的就是进行图像相减，可以找到边界点效果

（LK算法更好）