现代数字图像处理作业—对lena.bmp图像转成灰度图后，加入90%强度的椒盐噪声。尝试任何方法去恢复图像。

大家好，又见面了，我是你们的朋友全栈君。

作业代码及内容展示和分析：

1.代码

import cv2 as cv

import random as rd

import matplotlib.pyplot as plt

import numpy as np

le = cv.imread(‘lena.jpg’,0)

lena = le.copy()

def saltNoise_image(im,var): #生成椒盐噪声图函数

w,h = im.shape

saltNoiseNum = int(w*h*var) #噪声总的点数

all_xy_pos = [[[i,j] for j in range(h)]for i in range(w)]

xy = []

for i in all_xy_pos:

xy.extend(i)

salt_xy = rd.sample(xy,saltNoiseNum)

for i,j in salt_xy:

ra = rd.choice([0,1])

if ra == 0:

im[i,j] = 0

else:

im[i,j] = 255

return im

def border_image(im):

im[im==255] = 0

h,w = im.shape

im_mask = np.zeros((h+8,w+8),dtype=np.uint8)

im_mask[4:h+4,4:w+4] = im

return im_mask

def get_median(arr):

x = arr[arr!=0]

x.sort()

half = len(x)//2

if len(x)%2 == 0:

return x[half-1]//2 + x[half]//2

else:

return x[half]

def coreMat_cmt(a,b,im):

arr = im[a-1:a+2,b-1:b+2]

if arr.sum() == 0:

arr = im[a-2:a+3,b-2:b+3]

if arr.sum() == 0:

arr = im[a-3:a+4,b-3:b+4]

if arr.sum() == 0:

arr = im[a-4:a+5,b-4:b+5]

if arr.sum() == 0:

return 0

else:

gray = get_median(arr)

return gray

else:

gray = get_median(arr)

return gray

else:

gray = get_median(arr)

return gray

else:

gray = get_median(arr)

return gray

def run_filter(im):

h,w = im.shape

im_mask = np.zeros((h-8,w-8),dtype=np.uint8)

for i in range(4,h-4):

for j in range(4,w-4):

if im[i,j] ==0:

new_gray = coreMat_cmt(i,j,im)

im_mask[i-4,j-4] = new_gray

else:

im_mask[i-4,j-4] = im[i,j]

return im_mask

def filter_image(im):

for i in range(5):

border_im = border_image(im)

im = run_filter(border_im)

return im

noise_lena = saltNoise_image(lena,0.9)

filter_lena = filter_image(noise_lena)

cv.imshow(‘0’,le)

cv.imshow(‘1’,noise_lena)

cv.imshow(‘2’,filter_lena)

x = 0

for i in range(lena.shape[0]):

for j in range(lena.shape[1]):

x += abs(int(le[i,j]) – int(filter_lena[i,j]))

print(x/(400*400))

if cv.waitKey() == ord(‘A’):

cv.destroyAllWindows()

2.lena原图、90%强度噪声lena图像、恢复后图像；MSE为85.9

分析：

依据脉冲噪声的极值准则和不连续准则，利用D－S证据理论进行信息融合，在噪声修复阶段，对于检测窗口内非噪声点的中值，利用距离矩阵和偏差矩阵进行修正。

发布者：全栈程序员栈长，转载请注明出处：https://javaforall.cn/151642.html原文链接：https://javaforall.cn