关于图像融合论文中的常用技术指标（附加代码）

大家好，我是reedsways，最近大概做了三篇paper的工作，都和图像融合有关，因此，特意搞一个帖子来记录一下，那些常用的图像融合的量化指标！

首先，就是PSNR和SSIM的技术指标：

首先介绍一下：

直接上自用代码：

from tqdm import tqdm
import torch
import os
from torch.utils.data import Dataset
from torch.utils.data import DataLoader
import numpy as np
import cv2
from skimage.metrics import peak_signal_noise_ratio as psnr_loss
from skimage.metrics import structural_similarity as ssim_loss
import argparse



#直接在这里修改！！！！！！！！！！！
parser = argparse.ArgumentParser(description='PSNR SSIM script', add_help=False)
parser.add_argument('--input_images_path', default='./图片名字')
parser.add_argument('--image2smiles2image_save_path', default='./另一个图片的名字')

args = parser.parse_args()


def is_png_file(filename):
    return any(filename.endswith(extension) for extension in [".jpg", ".png", ".bmp"])


def load_img(filepath):
    img = cv2.cvtColor(cv2.imread(filepath), cv2.COLOR_BGR2RGB)
    img = img.astype(np.float32)
    img = img / 255.
    return img


class DataLoaderVal(Dataset):
    def __init__(self, target_transform=None):
        super(DataLoaderVal, self).__init__()

        self.target_transform = target_transform

        gt_dir = args.input_images_path
        input_dir = args.image2smiles2image_save_path

        clean_files = sorted(os.listdir(os.path.join(gt_dir)))
        noisy_files = sorted(os.listdir(os.path.join(input_dir)))

        self.clean_filenames = [os.path.join(gt_dir, x) for x in clean_files if is_png_file(x)]
        self.noisy_filenames = [os.path.join(input_dir, x) for x in noisy_files if is_png_file(x)]

        self.tar_size = len(self.clean_filenames)

    def __len__(self):
        return self.tar_size

    def __getitem__(self, index):
        tar_index = index % self.tar_size

        clean = torch.from_numpy(np.float32(load_img(self.clean_filenames[tar_index])))
        noisy = torch.from_numpy(np.float32(load_img(self.noisy_filenames[tar_index])))

        clean_filename = os.path.split(self.clean_filenames[tar_index])[-1]
        noisy_filename = os.path.split(self.noisy_filenames[tar_index])[-1]

        clean = clean.permute(2, 0, 1)
        noisy = noisy.permute(2, 0, 1)

        return clean, noisy, clean_filename, noisy_filename


def get_validation_data():
    return DataLoaderVal(None)


test_dataset = get_validation_data()
test_loader = DataLoader(dataset=test_dataset, batch_size=1, shuffle=False, num_workers=0, drop_last=False)

if __name__ == '__main__':

    psnr_val_rgb = []
    ssim_val_rgb = []
    for ii, data_test in enumerate(tqdm(test_loader), 0):
        rgb_groundtruth = data_test[0].numpy().squeeze().transpose((1, 2, 0))
        rgb_restored = data_test[1].cuda()

        rgb_restored = torch.clamp(rgb_restored, 0, 1).cpu().numpy().squeeze().transpose((1, 2, 0))
        psnr_val_rgb.append(psnr_loss(rgb_restored, rgb_groundtruth))
        ssim_val_rgb.append(ssim_loss(rgb_restored, rgb_groundtruth, multichannel=True))

    psnr_val_rgb = sum(psnr_val_rgb) / len(test_dataset)
    ssim_val_rgb = sum(ssim_val_rgb) / len(test_dataset)
    print("PSNR: %f, SSIM: %f " % (psnr_val_rgb, ssim_val_rgb))

接下来就是一些非参考的评价指标！！包括了！ Brenner、Laplacian、SMD、SMD2、Variance、Energy、Vollath and Entropy！

直接上代码（都是值越大越好）：

import cv2
import numpy as np
import math


# brenner梯度函数计算
def brenner(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    shape = np.shape(img)
    out = 0
    for x in range(0, shape[0] - 2):
        for y in range(0, shape[1]):
            out += (int(img[x + 2, y]) - int(img[x, y])) ** 2
    return out


# Laplacian梯度函数计算
def Laplacian(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    return cv2.Laplacian(img, cv2.CV_64F).var()


# SMD梯度函数计算
def SMD(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    shape = np.shape(img)
    out = 0
    for x in range(1, shape[0] - 1):
        for y in range(0, shape[1]):
            out += math.fabs(int(img[x, y]) - int(img[x, y - 1]))
            out += math.fabs(int(img[x, y] - int(img[x + 1, y])))
    return out


# SMD2梯度函数计算
def SMD2(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    shape = np.shape(img)
    out = 0
    for x in range(0, shape[0] - 1):
        for y in range(0, shape[1] - 1):
            out += math.fabs(int(img[x, y]) - int(img[x + 1, y])) * math.fabs(int(img[x, y] - int(img[x, y + 1])))
    return out


# 方差函数计算
def variance(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    out = 0
    u = np.mean(img)
    shape = np.shape(img)
    for x in range(0, shape[0]):
        for y in range(0, shape[1]):
            out += (img[x, y] - u) ** 2
    return out


# energy函数计算
def energy(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    shape = np.shape(img)
    out = 0
    for x in range(0, shape[0] - 1):
        for y in range(0, shape[1] - 1):
            out += ((int(img[x + 1, y]) - int(img[x, y])) ** 2) * ((int(img[x, y + 1] - int(img[x, y]))) ** 2)
    return out


# Vollath函数计算
def Vollath(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    shape = np.shape(img)
    u = np.mean(img)
    out = -shape[0] * shape[1] * (u ** 2)
    for x in range(0, shape[0] - 1):
        for y in range(0, shape[1]):
            out += int(img[x, y]) * int(img[x + 1, y])
    return out


# entropy函数计算
def entropy(img):
    '''
    :param img:narray 二维灰度图像
    :return: float 图像越清晰越大
    '''
    out = 0
    count = np.shape(img)[0] * np.shape(img)[1]
    p = np.bincount(np.array(img).flatten())
    for i in range(0, len(p)):
        if p[i] != 0:
            out -= p[i] * math.log(p[i] / count) / count
    return out


def main(img1):
    print('Brenner', brenner(img1) * 1e-6 )
    print('Laplacian', Laplacian(img1) * 1e-2 )
    print('SMD', SMD(img1) * 1e-5 )
    print('SMD2', SMD2(img1) * 1e-5 )
    print('Variance', variance(img1) * 1e-7 )
    print('Energy', energy(img1) * 1e-8 )
    print('Vollath', Vollath(img1) * 1e-7 )
    print('Entropy', entropy(img1))


if __name__ == '__main__':
    # 读入原始图像
    img1 = cv2.imread('input_images/name.jpg')
    # 灰度化处理
    img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
    main(img1)

最后最后，超级常用的四个参数！

信息熵！

import cv2
import numpy as np
import math

tmp = []
for i in range(256):
    tmp.append(0)
val = 0
k = 0
res = 0
# 'img/1-3.jpg'=6.0404 ;  out2.jpg=7.0361 ;result2=7.1585
image = cv2.imread('input_images/name.jpg', 0)
img = np.array(image)
for i in range(len(img)):
    for j in range(len(img[i])):
        val = img[i][j]
        tmp[val] = float(tmp[val] + 1)
        k = float(k + 1)
for i in range(len(tmp)):
    tmp[i] = float(tmp[i] / k)
for i in range(len(tmp)):
    if (tmp[i] == 0):
        res = res
    else:
        res = float(res - tmp[i] * (math.log(tmp[i]) / math.log(2.0)))
print(res)

均值和标准差！

from PIL import Image, ImageStat
import cv2
import numpy as np
import matplotlib.pyplot as plt

img = cv2.imread('input_images/name.jpg')
img = img.astype(np.float32) / 255
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# plt.imshow(img)

cv2.imshow('img', img)
# 通过img.copy()方法，复制img的数据到mean_img
mean_img = img.copy()
# 使用 .mean() 方法可得出 mean_img 的平均值
print(mean_img.mean())
# mean_img -= mean_img.mean() 等效于 mean_img = mean_img - mean_img.mean()
# 减去平均值，得出零平均值矩阵
mean_img -= mean_img.mean()
# 显示图像
# cv2.imshow(mean_img)
cv2.imshow('mean_img', mean_img)

std_img = mean_img.copy()
# 输出 std_img 的标准差
print(std_img.std())

# std_img /= std_img.mean() 等效于 std_img = std_img / std_img.mean()
# 除于标准差，得出单位方差矩阵
std_img /= std_img.std()
# 显示图像
# plt.imshow(std_img)
cv2.imshow('std_img', std_img)

cv2.waitKey(0)
cv2.destroyAllWindows()

平均梯度！

import cv2 as cv

import numpy as np

'''图像梯度(由x,y方向上的偏导数和偏移构成)，有一阶导数(sobel算子)和二阶导数(Laplace算子)
用于求解图像边缘，一阶的极大值，二阶的零点
一阶偏导在图像中为一阶差分，再变成算子(即权值)与图像像素值乘积相加，二阶同理
'''


def sobel_demo(image):
    grad_x = cv.Sobel(image, cv.CV_32F, 1, 0)  # 采用Scharr边缘更突出

    grad_y = cv.Sobel(image, cv.CV_32F, 0, 1)

    gradx = cv.convertScaleAbs(grad_x)  # 由于算完的图像有正有负，所以对其取绝对值

    grady = cv.convertScaleAbs(grad_y)

    # 计算两个图像的权值和，dst = src1alpha + src2beta + gamma

    gradxy = cv.addWeighted(gradx, 0.5, grady, 0.5, 0)

    cv.imshow("gradx", gradx)

    cv.imshow("grady", grady)

    cv.imshow("gradient", gradxy)


def laplace_demo(image):  # 二阶导数，边缘更细
    dst = cv.Laplacian(image, cv.CV_32F)

    lpls = cv.convertScaleAbs(dst)

    cv.imshow("laplace_demo", lpls)


def custom_laplace(image):
    # 以下算子与上面的Laplace_demo()是一样的，增强采用np.array([[1, 1, 1], [1, -8, 1], [1, 1, 1]])kernel = np.array([[1, 1, 1], [1, -8, 1], [1, 1, 1]])

    dst = cv.filter2D(image, cv.CV_32F, kernel=kernel)

    lpls = cv.convertScaleAbs(dst)

    cv.imshow("custom_laplace", lpls)


def Scharr(img):
    scharrx = cv.Scharr(img, cv.CV_64F, dx=1, dy=0)
    scharrx = cv.convertScaleAbs(scharrx)

    scharry = cv.Scharr(img, cv.CV_64F, dx=0, dy=1)
    scharry = cv.convertScaleAbs(scharry)

    result = cv.addWeighted(scharrx, 0.5, scharry, 0.5, 0)

    cv.imshow("scharrx", scharrx)
    cv.imshow("scharry", scharry)
    cv.imshow("result", result)


src = cv.imread("input_images/name.jpg")

cv.imshow("original", src)

sobel_demo(src)

laplace_demo(src)

# custom_laplace(src)
Scharr(src)

cv.waitKey(0)  # 等有键输入或者1000ms后自动将窗口消除，0表示只用键输入结束窗口

cv.destroyAllWindows()  # 关闭所有窗口