图像质量评估指标——PSNR介绍及计算方法
PSNR全称为Peak Signal to Noise Ratio,即峰值信噪比,是一种评价图像的客观标准,经常用于计算两幅图像的可视化误差,如衡量原图与经过压缩后图像之间的可视化差异、生成网络生成的图像与真实图像之间的差异等等。
计算方法
假设处理后图像为
I
I
I、真实图像为
K
K
K,则峰值噪声比定义为:
P
S
N
R
=
10
⋅
log
10
(
M
A
X
2
M
S
E
)
=
20
⋅
log
10
(
M
A
X
M
S
E
)
PSNR=10·\log_{10}(\frac{MAX^2}{MSE})=20·\log_{10}(\frac{MAX}{\sqrt{MSE}})
PSNR=10⋅log10(MSEMAX2)=20⋅log10(MSEMAX)
其中,
M
A
X
MAX
MAX表示图像上像素点的最大数值,常定义为
1
1
1或者
255
255
255,
M
S
E
MSE
MSE定义为:
M
S
E
=
1
m
n
∑
i
=
0
m
−
1
∑
j
=
0
n
−
1
∣
∣
I
(
i
,
j
)
−
K
(
i
,
j
)
∣
∣
2
MSE=\frac{1}{mn}\sum^{m-1}_{i=0}\sum^{n-1}_{j=0}||I(i,j)-K(i,j)||^2
MSE=mn1i=0∑m−1j=0∑n−1∣∣I(i,j)−K(i,j)∣∣2
代码实现
- 参考论文《Spatial Attentive Single-Image Deraining with a High Quality Real Rain Dataset》的源码:https://github.com/stevewongv/SPANet
# 这里以生成网络中测试模型性能为例,所以pred与gt均是tensor格式
# 并且在模型训练过程中,图像数据均被归一化至0到1之间,因此max默认1.0
def psnr(pred, gt, max=1.0):
# 需要先切断梯度,将数据转为numpy格式进行计算
pred = pred.clamp(0, max).cpu().detach().numpy()
gt = gt.clamp(0, max).cpu().detach().numpy()
# 两个图像数值做差,之后求根下MSE
imdff = pred - gt
rmse = math.sqrt(np.mean(imdff ** 2))
# 如果像素完全相同,即rmse为0,则PSNR直接返回100,防止后续计算log时报错
if rmse == 0:
return 100
# 按公式计算
return 20 * math.log10(max / rmse)
直接调用skimage库中的函数
from skimage.metrics import peak_signal_noise_ratio
# 这里data_range表示图像像素的最大值,即公式中的MAX
# 如果不传入data_range,则程序默认选择image_true中最大的像素值
# 或者最大值减最小值(如果像素点存在负数情况),具体可见源码
psnr = peak_signal_noise_ratio(image_true, image_test, data_range=None)
skimage中计算PSNR的源码
def peak_signal_noise_ratio(image_true, image_test, *, data_range=None):
"""
Compute the peak signal to noise ratio (PSNR) for an image.
Parameters
----------
image_true : ndarray
Ground-truth image, same shape as im_test.
image_test : ndarray
Test image.
data_range : int, optional
The data range of the input image (distance between minimum and
maximum possible values). By default, this is estimated from the image
data-type.
Returns
-------
psnr : float
The PSNR metric.
Notes
-----
.. versionchanged:: 0.16
This function was renamed from ``skimage.measure.compare_psnr`` to
``skimage.metrics.peak_signal_noise_ratio``.
References
----------
.. [1] https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio
"""
check_shape_equality(image_true, image_test)
# 如果不指定data_range的话,则默认取image_true中的最大值,或者最大值减最小值
if data_range is None:
if image_true.dtype != image_test.dtype:
warn("Inputs have mismatched dtype. Setting data_range based on "
"image_true.")
dmin, dmax = dtype_range[image_true.dtype.type]
true_min, true_max = np.min(image_true), np.max(image_true)
if true_max > dmax or true_min < dmin:
raise ValueError(
"image_true has intensity values outside the range expected "
"for its data type. Please manually specify the data_range.")
if true_min >= 0:
# most common case (255 for uint8, 1 for float)
data_range = dmax
else:
data_range = dmax - dmin
# 将数据转变为float类型
image_true, image_test = _as_floats(image_true, image_test)
# 计算MSE
err = mean_squared_error(image_true, image_test)
# 进一步计算PSNR
return 10 * np.log10((data_range ** 2) / err)
以上仅是笔者个人见解,若有问题,欢迎指正。