非局部均值滤波NL_means--python实现

import numpy as np
import matplotlib.pyplot as plt
from scipy import signal
from skimage.restoration import estimate_sigma
from skimage import measure, util
from skimage.io import imread, imsave
import cv2 as cv

img = cv.imread("C:/Users/W1998/Desktop/photo/10m_plasticbag_1583.jpg", 0)
img_noisy = cv.imread("C:/Users/W1998/Desktop/photo/noisy.jpg", 0)
#导入的两张图片的大小必须一致

#print(img_noisy.shape)  #查看原图的尺寸大小
#print(img.shape)   #查看加噪图像的尺寸大小（我用的是高斯噪声）
#img_n = cv.resize(img_noisy, (850,491), interpolation=cv.INTER_CUBIC)#将上面两个图像尺寸大小缩放一致

def clip(im):
    return np.maximum(0.,np.minimum(1.,im))

from os.path import normpath as fn # Fixes window/linux path conventions
import warnings
warnings.filterwarnings('ignore')

lena = np.float32(img)/255.
noisy_lena = np.float32(img_n)/255.

# Gaussian Filtering高斯滤波去噪
def gaussian(X,K):
    x, y = np.mgrid[-K:K + 1, -K:K + 1]
    g = np.exp(-(x ** 2 / float(K) + y ** 2 / float(K)))
    kern = g / g.sum()

    new_im = signal.convolve2d(X,kern,mode='same')
    return new_im


def wavelet(X,levels, lmain):
    def im2wv(img, nLev):
        # pyr array
        pyr = []
        h_mat = np.array([[1, 1, 1, 1],
                          [-1, 1, -1, 1],
                          [-1, -1, 1, 1],
                          [1, -1, -1, 1]])
        for i in range(nLev):
            n, mid = len(img), len(img) // 2
            # split image up for HWT
            a = img[:n:2, :n:2]
            b = img[1:n:2, :n:2]
            c = img[:n:2, 1:n:2]
            d = img[1:n:2, 1:n:2]
            vec = np.array([a, b, c, d])
            # reshape vector to perform mat mult
            D = 1 / 2 * np.dot(h_mat, vec.reshape(4, mid * mid))
            L, H1, H2, H3 = D.reshape([4, mid, mid])
            pyr.append([H1, H2, H3])
            img = L
        pyr.append(L)
        return pyr

    def wv2im(pyr):
        # need inverse of unitary matrix
        h_mat = np.array([[1, 1, 1, 1],
                          [-1, 1, -1, 1],
                          [-1, -1, 1, 1],
                          [1, -1, -1, 1]])
        h_mat_inv = np.linalg.inv(h_mat)
        # last spot in pyramid is small img
        # iterate in reverse to reconstruct
        L = pyr[-1]
        for [H1, H2, H3] in reversed(pyr[:-1]):
            n, n2 = len(L), len(L) * 2
            vec = np.array([L, H1, H2, H3])
            # reshape vector to perform mat mult
            D = 2 * np.dot(h_mat_inv, vec.reshape(4, n * n))
            a, b, c, d = D.reshape([4, n, n])
            # assign pixels to correct spots in img
            img = np.empty((n2, n2))
            img[:n2:2, :n2:2] = a
            img[1:n2:2, :n2:2] = b
            img[:n2:2, 1:n2:2] = c
            img[1:n2:2, 1:n2:2] = d
            L = img
        return L

    # Return corresponding coefficients x (same shape/size)
    # that minimizes (x - y)^2 + lmbda * abs(x)
    def denoise_coeff(y, lmbda):
        x = np.copy(y)
        x[np.where(y > lmbda / 2.0)] -= lmbda / 2.0
        x[np.where(y < -lmbda / 2.0)] += lmbda / 2.0
        x[np.where(np.logical_and(y>-lmbda/2.0,y