之前做实验发现了一个问题,好像用python的skimage.metrics包计算PSNR等指标和matlab里面算出来的不太一样,当时全转成python算了,现在考虑到又有不同的对比实验啥的,打算统一一下python和matlab的指标计算:
(垃圾csdn,自动把我的文章vip可见了,真该死啊,现在调回公开了)
代码
python版本:
from skimage.metrics import peak_signal_noise_ratio as psnr,structural_similarity as ssim,mean_squared_error as mse
from scipy.io import loadmat
import numpy as np
# 读取图片
x0 = loadmat('A.mat')["data"]
x2 = loadmat('AA.mat')["data"]
x0 = np.squeeze(x0)
x2 = np.squeeze(x2)
# 归一化
maxvalue1=x0.max()
minvalue1=x0.min()
x0 = (x0 - minvalue1) / (maxvalue1 - minvalue1)
maxvalue1=x2.max()
minvalue1=x2.min()
x2 = (x2 - minvalue1) / (maxvalue1 - minvalue1)
psnr0 = psnr(x0, x2, data_range=1)
ssim0 = ssim(x0,x2,gaussian_weights=True, use_sample_covariance=False, data_range=1.0)
mse0 = mse(x0,x2)
print(psnr0)
print(ssim0)
print(mse0)
matlab版本:
clear
A = load('A.mat').data;
B = load('AA.mat').data;
A = squeeze(A);
B = squeeze(B);
X = double(A);
Y = double(B);
maxvalue1=max(max(A));
minvalue1=min(min(A));
A = (A - minvalue1) / (maxvalue1 - minvalue1);
maxvalue1=max(max(B));
minvalue1=min(min(B));
B = (B - minvalue1) / (maxvalue1 - minvalue1);
[psnr,mse]= psnr_zl(A,B,1)
ssim1 = ssim_zl(A,B)
function [PSNR, MSE] = psnr_zl(X, Y,datarange) % 计算峰值信噪比PSNR
MSE = mean(mean((X-Y).^2));
PSNR = 10 * log10((datarange^2) / MSE);
end
function [mssim, ssim_map] = ssim_zl(img1, img2, K, window, L)
%前两个输入是两张图片,K,L是计算均值、标准差、C1、C2需要的参数,windows是滤波器窗口大小
if (nargin < 2 || nargin > 5)
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
if (size(img1) ~= size(img2))
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
[M N] = size(img1);
if (nargin == 2)
if ((M < 11) || (N < 11)) % 图像大小过小,则没有意义。
% ssim_index = -Inf;
ssim_map = -Inf;
return
end
window = fspecial('gaussian', 11, 1.5); % 参数一个标准偏差1.5,11*11的高斯低通滤波。
K(1) = 0.01; % default settings
K(2) = 0.03; %
L = 255; %
end
if (nargin == 3)
if ((M < 11) || (N < 11))
% ssim_index = -Inf;
ssim_map = -Inf;
return
end
window = fspecial('gaussian', 11, 1.5);
L = 255;
if (length(K) == 2)
if (K(1) < 0 || K(2) < 0)
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
else
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
end
if (nargin == 4)
[H W] = size(window);
if ((H*W) < 4 || (H > M) || (W > N))
% ssim_index = -Inf;
ssim_map = -Inf;
return
end
L = 255;
if (length(K) == 2)
if (K(1) < 0 || K(2) < 0)
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
else
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
end
if (nargin == 5)
[H W] = size(window);
if ((H*W) < 4 || (H > M) || (W > N))
% ssim_index = -Inf;
ssim_map = -Inf;
return
end
if (length(K) == 2)
if (K(1) < 0 || K(2) < 0)
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
else
% ssim_index = -Inf;
ssim_map = -Inf;
return;
end
end
L=1;
C1 = (K(1)*L)^2; % 计算C1参数,给亮度L(x,y)用。
C2 = (K(2)*L)^2; % 计算C2参数,给对比度C(x,y)用。
window = window/sum(sum(window)); %滤波器归一化操作。
img1 = double(img1);
img2 = double(img2);
mu1 = filter2(window, img1, 'valid'); % 对图像进行滤波因子加权
mu2 = filter2(window, img2, 'valid'); % 对图像进行滤波因子加权
mu1_sq = mu1.*mu1; % 计算出Ux平方值。
mu2_sq = mu2.*mu2; % 计算出Uy平方值。
mu1_mu2 = mu1.*mu2; % 计算Ux*Uy值。
sigma1_sq = filter2(window, img1.*img1, 'valid') - mu1_sq; % 计算sigmax (标准差)
sigma2_sq = filter2(window, img2.*img2, 'valid') - mu2_sq; % 计算sigmay (标准差)
sigma12 = filter2(window, img1.*img2, 'valid') - mu1_mu2; % 计算sigmaxy(标准差)
if (C1 > 0 && C2 > 0)
ssim_map = ((2*mu1_mu2 + C1).*(2*sigma12 + C2))./((mu1_sq + mu2_sq + C1).*(sigma1_sq + sigma2_sq + C2));
else
numerator1 = 2*mu1_mu2 + C1;
numerator2 = 2*sigma12 + C2;
denominator1 = mu1_sq + mu2_sq + C1;
denominator2 = sigma1_sq + sigma2_sq + C2;
ssim_map = ones(size(mu1));
index = (denominator1.*denominator2 > 0);
ssim_map(index) = (numerator1(index).*numerator2(index))./(denominator1(index).*denominator2(index));
index = (denominator1 ~= 0) & (denominator2 == 0);
ssim_map(index) = numerator1(index)./denominator1(index);
end
mssim = mean2(ssim_map);
return
end
输出结果:
python:
matlab:
可以看到结果是一样的。
另外,那个matlab代码是网上copy的,可能有点冗余(比如那个L,对应python的datarange,我强行设置成1了),有兴趣的可以优化下。
使用的图片大小为512*512,归一化值域到[0,1],非自然图像,是单通道的CT图。
自然图像也许也能用,但我懒得试了,试了的人可以评论一下。