中值滤波
中值滤波的原理：
中值滤波，其原理在于使用卷积核内的中值来代替中心点的值。例如，当一个点值为255时，而周围其他点全都低于120，这时候它就是一个明显的噪点。使用中值滤波时，就会用其他的点代替该点的值。不敢说替换的点就是真值，但是它一定比原来的点更接近真值。
下面将介绍如何使用python代码来实现。（文中所使用的图像为网图，如有侵权请及时联系删除）
from PIL import Image
import numpy as np
class Mdiean_Filter:
def __init__(self, source_img):
self.source_img = source_img #原图
self.noise_img = './salt_noise.jpg' #加了椒盐噪声之后的图像
self.mdiean_img = './mdiean_filter.jpg' #中值滤波的图像
self.k = 3
def Add_Salt_Noise(self): # 加椒盐噪声
img = Image.open(self.source_img)
imgarray = np.array(img)
height,width = imgarray.shape[0], imgarray.shape[1]
for i in range(height):
for j in range(width):
if np.random.random(1) < 0.05:
if np.random.random(1) < 0.3:
imgarray[i][j] = 0
else:
imgarray[i][j] = 255
new_img = Image.fromarray(imgarray)
new_img.save(self.noise_img)
return imgarray
def Mdiean_Filtering(self, padding = None): #中值滤波会变为二值图
img = Image.open(self.noise_img)
imgarray = np.array(img)
height, width = imgarray.shape[0], imgarray.shape[1]
print(imgarray.shape)
if not padding:
edge = int((self.k -1)/2)
if height -1 -edge <=edge or width -1-edge<=edge:
print("the kenerl is to long")
return None
new_arr = np.zeros((height, width,3), dtype='uint8')
print(new_arr.shape)
for i in range(height):
for j in range(width):
if i <=edge -1 or i >= height -1 -edge or j <=edge -1 or j >= height -1 -edge:
new_arr[i, j] = imgarray[i, j]
else:
new_arr[i, j] = np.median(imgarray[i - edge:i + edge + 1, j -edge:j+edge + 1]) #numpy的代码计算时会把矩阵三个数值变成一样，这也是色彩损失的原因
new_img = Image.fromarray(new_arr)
new_img.save(self.mdiean_img)
def Mdiean_Filter1(self, padding = None): #中值滤波不损失色彩
imgarray = self.Add_Salt_Noise()
height, width = imgarray.shape[0], imgarray.shape[1]
if not padding:
edge = int((self.k -1)/2)
if height -1 -edge <=edge or width -1-edge<=edge:
print("the kenerl is to long")
return None
for i in range(height):
for j in range(width):
if i <=edge -1 or i >= height -1 -edge or j <=edge -1 or j >= height -1 -edge:
imgarray[i][j] = imgarray[i][j]
else:
num = []
for m in range(i - edge, i + edge + 1):
for n in range(j -edge, j+edge + 1):
num.append((imgarray[m][n])[0]) #这里通过彩色图像第一个值计算中值也可以改为第二个或者第三个
temp = np.median(num)
idex_tem = num.index(temp) #获取中值在数组中的坐标
l1 = int((idex_tem / self.k ))- edge + i #根据进制转换反推出中值在图像中的坐标
l2 = (idex_tem % self.k )- edge + j
print(imgarray[l1][l2])
imgarray[i][j] = imgarray[l1][l2] #赋值
#num1 = np.sort(num)[int(self.k * self.k / 2)]
#print(idex_tem)
#temp = np.median(imgarray[i][j], imgarray[i - edge][j], imgarray[i - edge][j])
#imgarray[i][j] = np.median(imgarray[i - edge:i + edge + 1, j -edge:j+edge + 1])
new_img = Image.fromarray(imgarray)
new_img.save(self.mdiean_img)
if __name__=="__main__":
example = 'example.jpg'
filter = Mdiean_Filter(example)
#filter.Add_Salt_Noise()
filter.Mdiean_Filter1()