Image denoising using nlmeas

Use nlmeas Denoise the image

Effect display

Code implementation

# Use nlmeans Methods the image was denoised 

import cv2
import numpy as np
import matplotlib.pyplot as plt

import skimage
from skimage import data,img_as_float
from skimage.restoration import denoise_nl_means,estimate_sigma
from skimage.metrics import  peak_signal_noise_ratio
from skimage.util import  random_noise

# Use  skimage  It's from Ku nlmeans  The algorithm operates 
def nlmeansSkimage(src):
    # Load the picture first 
    img=skimage.io.imread(src)
    img=skimage.img_as_float(img)

    # Artificially add noise to the picture , You can delete 
    sigma=0.08
    noisy=random_noise(img,var=sigma**2)

    # Estimate the standard deviation of the noise in the current input picture 
    sigma_est=np.mean(estimate_sigma(noisy,channel_axis=-1))
    print(f'the estimate Image noise is{
      sigma_est}')

    # take pathch  Some attributes of the block are fixed 
    patch_kw=dict(
            patch_size=5,
            patch_distance=6,
            channel_axis=-1
    )
#denoise_nl_means(image, patch_size=7, patch_distance=11, h=0.1, multichannel=False, fast_mode=True, sigma=0.0,, preserve_range=False, channel_axis=None)
    #image= The denoised image to be input , It can be a grayscale image , Can also make rgb chart 
    #patch_size: For denoising patch Size 
    #patch_distance: Search for patch Maximum pixel distance 
    #h: Cut off distance ,h The higher the value , Accept patch The higher the degree of freedom , At the same time, the smoother the image , But the boundary will also become blurred . For standard deviation sigma Gaussian noise of , Experience is better than sigma Less valuable 
    #multichannel/channel_axis: Specifies whether the input image is multichannel , If not specified, it defaults to grayscale 
    #fastmode: Specify whether to use  nlmeans Fast version of 
    #sigma: Standard deviation of Gaussian noise , If provided, calculate pach The weight will take into account the expected noise variance 
    #preserve_range: Whether to keep the range of original values , otherwise , It's like Korean dramas img_as_float Rule transformation input u Image 

    denoiseNlmeans=denoise_nl_means(noisy,h=1.15*sigma_est,fast_mode=False,**patch_kw)

    # Provide predictions sigma Estimated value 
    denoiseNlmeansSigma=denoise_nl_means(noisy,h=0.8*sigma_est,sigma=sigma_est,
                                        fast_mode=False,**patch_kw)

    # Use fast nlmeans Method 
    denoiseNlmeansFast=denoise_nl_means(noisy,h=0.8*sigma_est,
                                        fast_mode=True,**patch_kw)

    # Use fast nlmeans And combine sigma_est
    denoiseNlmeansFastSigma=denoise_nl_means(noisy,h=0.8*sigma_est,sigma=sigma_est,
                                        fast_mode=True,**patch_kw)


    # Of the calculated image PSNR
    psnr_noisy=peak_signal_noise_ratio(img,noisy)
    psnr_denoiseNlmeans=peak_signal_noise_ratio(img,denoiseNlmeans)
    psnr_denoiseNlmeansSigma=peak_signal_noise_ratio(img,denoiseNlmeansSigma)
    psnr_denoiseNlmeansFast=peak_signal_noise_ratio(img,denoiseNlmeansFast)
    psnr_denoiseNlmeansFastSigma=peak_signal_noise_ratio(img,denoiseNlmeansFastSigma)





    # Draw the output image 

    fig,ax=plt.subplots(nrows=2, ncols=3, figsize=(8,6),
                       sharex=True,sharey=True)
    ax[0,0].imshow(noisy)
    ax[0,0].axis('off')
    ax[0,0].set_title(f'noisy Image\nPSNR={
      psnr_noisy}')

    ax[0, 1].imshow(denoiseNlmeans)
    ax[0, 1].axis('off')
    ax[0, 1].set_title(f'non-local means\n(slow,PSNR={
      psnr_denoiseNlmeans})')

    ax[0, 2].imshow(denoiseNlmeansSigma)
    ax[0, 2].axis('off')
    ax[0, 2].set_title('non-local means\n(slow,using $\\sigma_{est}$)'+f'PSNR={
      psnr_denoiseNlmeansSigma}')

    ax[1, 0].imshow(img)
    ax[1, 0].axis('off')
    ax[1, 0].set_title('Original Image')

    ax[1, 1].imshow(denoiseNlmeansFast)
    ax[1, 1].axis('off')
    ax[1, 1].set_title(f'non-local means\n(fast)PSNR={
      psnr_denoiseNlmeansFast}')

    ax[1, 2].imshow(denoiseNlmeansFastSigma)
    ax[1, 2].axis('off')
    ax[1, 2].set_title('non-local means\n(fast,using $\\sigma_{est}$)'+f'PSNR={
      psnr_denoiseNlmeansFastSigma}')

    fig.tight_layout()
    plt.show()