马鞍山住房建设委员会网站优化方案丛书官网

图像融合评估指标Python版

这篇博客利用Python把大部分图像融合指标基于图像融合评估指标复现了，从而方便大家更好的使用Python进行指标计算，以及一些I/O 操作。除了几个特征互信息的指标没有成功复现之外，其他指标均可以通过这篇博客提到的Python程序计算得到，其中SSIM和MS_SSIM是基于PyTorch实现的可能无法与原来的程序保持一致，同时使用了一些矩阵运算加速了Nabf和Qabf的计算。但不幸的是在计算VIF时设计大量的卷积运算，而博主在Python中采用cipy.signal.convolve2d来替换MATLAB中的filter函数,导致时间消耗较大，如果你不需要计算VIF可以直接注释掉相关语句 并设置VIF=1即可。

在原来的MATLAB程序中由于没有充分考虑数据类型的影响，在计算SD是会由于uint8数据类型的限制，但是部分数据被截断，在Python中已经解决了这个Bug，同时也在原来的MATLAB版本中修正了这个问题。

在Python版的程序中，只有计算EN和MI是使用的是int型数据，其他指标均使用float型数据。此外除了计算MSE和PSNR时将数据归一化到[0,1]之外，计算其他指标时，数据范围均为[0,255]。

评估指标	缩写
信息熵	EN
空间频率	SF
标准差	SD
峰值信噪比	PSNR
均方误差	MSE
互信息	MI
视觉保真度	VIF
平均梯度	AG
相关系数	CC
差异相关和	SCD
基于梯度的融合性能	Qabf
结构相似度测量	SSIM
多尺度结构相似度测量	MS-SSIM
基于噪声评估的融合性能	Nabf

性能评估指标主要分为四类，分别是基于信息论的评估指标，主要包括** EN、MI、PSNR**、基于结构相似性的评估指标，主要包括SSIM、MS_SSIM、MSE、基于图像特征的评估指标， 主要包括SF、SD、AG，基于人类视觉感知的评估指标，主要包括VIF、以及基于源图像与生成图像的评估指标，主要包括CC、SCD、Qabf、Nabf。

接下来是部分程序：

单张图像测试程序: eval_one_image.py

from PIL import Image
from Metric import *
from time import time
import warnings
warnings.filterwarnings("ignore")def evaluation_one(ir_name, vi_name, f_name):f_img = Image.open(f_name).convert('L')ir_img = Image.open(ir_name).convert('L')vi_img = Image.open(vi_name).convert('L')f_img_int = np.array(f_img).astype(np.int32)f_img_double = np.array(f_img).astype(np.float32)ir_img_int = np.array(ir_img).astype(np.int32)ir_img_double = np.array(ir_img).astype(np.float32)vi_img_int = np.array(vi_img).astype(np.int32)vi_img_double = np.array(vi_img).astype(np.float32)EN = EN_function(f_img_int)MI = MI_function(ir_img_int, vi_img_int, f_img_int, gray_level=256)SF = SF_function(f_img_double)SD = SD_function(f_img_double)AG = AG_function(f_img_double)PSNR = PSNR_function(ir_img_double, vi_img_double, f_img_double)MSE = MSE_function(ir_img_double, vi_img_double, f_img_double)VIF = VIF_function(ir_img_double, vi_img_double, f_img_double)CC = CC_function(ir_img_double, vi_img_double, f_img_double)SCD = SCD_function(ir_img_double, vi_img_double, f_img_double)Qabf = Qabf_function(ir_img_double, vi_img_double, f_img_double)Nabf = Nabf_function(ir_img_double, vi_img_double, f_img_double)SSIM = SSIM_function(ir_img_double, vi_img_double, f_img_double)MS_SSIM = MS_SSIM_function(ir_img_double, vi_img_double, f_img_double)return EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIMif __name__ == '__main__':f_name = r'E:\Desktop\metric\Test\Results\TNO\GTF\01.png'ir_name = r'E:\Desktop\metric\Test\datasets\TNO\ir\01.png'vi_name = r'E:\Desktop\metric\Test\datasets\TNO\vi\01.png'EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIM = evaluation_one(ir_name, vi_name, f_name)print('EN:', round(EN, 4))print('MI:', round(MI, 4))print('SF:', round(SF, 4))print('AG:', round(AG, 4))print('SD:', round(SD, 4))print('CC:', round(CC, 4))print('SCD:', round(SCD, 4))print('VIF:', round(VIF, 4))print('MSE:', round(MSE, 4))print('PSNR:', round(PSNR, 4))print('Qabf:', round(Qabf, 4))print('Nabf:', round(Nabf, 4))print('SSIM:', round(SSIM, 4))print('MS_SSIM:', round(MS_SSIM, 4))

测试一个方法中所有图像指标的程序: eval_one_method.py

import numpy as np
from PIL import Image
from Metric import *
from natsort import natsorted
from tqdm import tqdm
import os
import statistics
import warnings
from openpyxl import Workbook, load_workbook
from openpyxl.utils import get_column_letter
warnings.filterwarnings("ignore")def write_excel(excel_name='metric.xlsx', worksheet_name='VIF', column_index=0, data=None):try:workbook = load_workbook(excel_name)except FileNotFoundError:# 文件不存在，创建新的 Workbookworkbook = Workbook()# 获取或创建一个工作表if worksheet_name in workbook.sheetnames:worksheet = workbook[worksheet_name]else:worksheet = workbook.create_sheet(title=worksheet_name)# 在指定列中插入数据column = get_column_letter(column_index + 1)for i, value in enumerate(data):cell = worksheet[column + str(i+1)]cell.value = value# 保存文件workbook.save(excel_name)def evaluation_one(ir_name, vi_name, f_name):f_img = Image.open(f_name).convert('L')ir_img = Image.open(ir_name).convert('L')vi_img = Image.open(vi_name).convert('L')f_img_int = np.array(f_img).astype(np.int32)f_img_double = np.array(f_img).astype(np.float32)ir_img_int = np.array(ir_img).astype(np.int32)ir_img_double = np.array(ir_img).astype(np.float32)vi_img_int = np.array(vi_img).astype(np.int32)vi_img_double = np.array(vi_img).astype(np.float32)EN = EN_function(f_img_int)MI = MI_function(ir_img_int, vi_img_int, f_img_int, gray_level=256)SF = SF_function(f_img_double)SD = SD_function(f_img_double)AG = AG_function(f_img_double)PSNR = PSNR_function(ir_img_double, vi_img_double, f_img_double)MSE = MSE_function(ir_img_double, vi_img_double, f_img_double)VIF = VIF_function(ir_img_double, vi_img_double, f_img_double)CC = CC_function(ir_img_double, vi_img_double, f_img_double)SCD = SCD_function(ir_img_double, vi_img_double, f_img_double)Qabf = Qabf_function(ir_img_double, vi_img_double, f_img_double)Nabf = Nabf_function(ir_img_double, vi_img_double, f_img_double)SSIM = SSIM_function(ir_img_double, vi_img_double, f_img_double)MS_SSIM = MS_SSIM_function(ir_img_double, vi_img_double, f_img_double)return EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIMif __name__ == '__main__':with_mean = TrueEN_list = []MI_list = []SF_list = []AG_list = []SD_list = []CC_list = []SCD_list = []VIF_list = []MSE_list = []PSNR_list = []Qabf_list = []Nabf_list = []SSIM_list = []MS_SSIM_list = []filename_list = ['']dataset_name = 'test_imgs'ir_dir = os.path.join('..\datasets', dataset_name, 'ir')vi_dir = os.path.join('..\datasets', dataset_name, 'vi')Method = 'SeAFusion'f_dir = os.path.join('..\Results', dataset_name, Method)save_dir = '..\Metric'os.makedirs(save_dir, exist_ok=True)metric_save_name = os.path.join(save_dir, 'metric_{}_{}.xlsx'.format(dataset_name, Method))filelist = natsorted(os.listdir(ir_dir))eval_bar = tqdm(filelist)for _, item in enumerate(eval_bar):ir_name = os.path.join(ir_dir, item)vi_name = os.path.join(vi_dir, item)f_name = os.path.join(f_dir, item)EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIM = evaluation_one(ir_name, vi_name, f_name)EN_list.append(EN)MI_list.append(MI)SF_list.append(SF)AG_list.append(AG)SD_list.append(SD)CC_list.append(CC)SCD_list.append(SCD)VIF_list.append(VIF)MSE_list.append(MSE)PSNR_list.append(PSNR)Qabf_list.append(Qabf)Nabf_list.append(Nabf)SSIM_list.append(SSIM)MS_SSIM_list.append(MS_SSIM)filename_list.append(item)eval_bar.set_description("{} | {}".format(Method, item))if with_mean:# 添加均值EN_list.append(np.mean(EN_list))MI_list.append(np.mean(MI_list))SF_list.append(np.mean(SF_list))AG_list.append(np.mean(AG_list))SD_list.append(np.mean(SD_list))CC_list.append(np.mean(CC_list))SCD_list.append(np.mean(SCD_list))VIF_list.append(np.mean(VIF_list))MSE_list.append(np.mean(MSE_list))PSNR_list.append(np.mean(PSNR_list))Qabf_list.append(np.mean(Qabf_list))Nabf_list.append(np.mean(Nabf_list))SSIM_list.append(np.mean(SSIM_list))MS_SSIM_list.append(np.mean(MS_SSIM_list))filename_list.append('mean')## 添加标准差EN_list.append(np.std(EN_list))MI_list.append(np.std(MI_list))SF_list.append(np.std(SF_list))AG_list.append(np.std(AG_list))SD_list.append(np.std(SD_list))CC_list.append(np.std(CC_list[:-1]))SCD_list.append(np.std(SCD_list))VIF_list.append(np.std(VIF_list))MSE_list.append(np.std(MSE_list))PSNR_list.append(np.std(PSNR_list))Qabf_list.append(np.std(Qabf_list))Nabf_list.append(np.std(Nabf_list))SSIM_list.append(np.std(SSIM_list))MS_SSIM_list.append(np.std(MS_SSIM_list))filename_list.append('std')## 保留三位小数EN_list = [round(x, 3) for x in EN_list]MI_list = [round(x, 3) for x in MI_list]SF_list = [round(x, 3) for x in SF_list]AG_list = [round(x, 3) for x in AG_list]SD_list = [round(x, 3) for x in SD_list]CC_list = [round(x, 3) for x in CC_list]SCD_list = [round(x, 3) for x in SCD_list]VIF_list = [round(x, 3) for x in VIF_list]MSE_list = [round(x, 3) for x in MSE_list]PSNR_list = [round(x, 3) for x in PSNR_list]Qabf_list = [round(x, 3) for x in Qabf_list]Nabf_list = [round(x, 3) for x in Nabf_list]SSIM_list = [round(x, 3) for x in SSIM_list]MS_SSIM_list = [round(x, 3) for x in MS_SSIM_list]EN_list.insert(0, '{}'.format(Method))MI_list.insert(0, '{}'.format(Method))SF_list.insert(0, '{}'.format(Method))AG_list.insert(0, '{}'.format(Method))SD_list.insert(0, '{}'.format(Method))CC_list.insert(0, '{}'.format(Method))SCD_list.insert(0, '{}'.format(Method))VIF_list.insert(0, '{}'.format(Method))MSE_list.insert(0, '{}'.format(Method))PSNR_list.insert(0, '{}'.format(Method))Qabf_list.insert(0, '{}'.format(Method))Nabf_list.insert(0, '{}'.format(Method))SSIM_list.insert(0, '{}'.format(Method))MS_SSIM_list.insert(0, '{}'.format(Method))write_excel(metric_save_name, 'EN', 0, filename_list)write_excel(metric_save_name, "MI", 0, filename_list)write_excel(metric_save_name, "SF", 0, filename_list)write_excel(metric_save_name, "AG", 0, filename_list)write_excel(metric_save_name, "SD", 0, filename_list)write_excel(metric_save_name, "CC", 0, filename_list)write_excel(metric_save_name, "SCD", 0, filename_list)write_excel(metric_save_name, "VIF", 0, filename_list)write_excel(metric_save_name, "MSE", 0, filename_list)write_excel(metric_save_name, "PSNR", 0, filename_list)write_excel(metric_save_name, "Qabf", 0, filename_list)write_excel(metric_save_name, "Nabf", 0, filename_list)write_excel(metric_save_name, "SSIM", 0, filename_list)write_excel(metric_save_name, "MS_SSIM", 0, filename_list)write_excel(metric_save_name, 'EN', 1, EN_list)write_excel(metric_save_name, 'MI', 1, MI_list)write_excel(metric_save_name, 'SF', 1, SF_list)write_excel(metric_save_name, 'AG', 1, AG_list)write_excel(metric_save_name, 'SD', 1, SD_list)write_excel(metric_save_name, 'CC', 1, CC_list)write_excel(metric_save_name, 'SCD', 1, SCD_list)write_excel(metric_save_name, 'VIF', 1, VIF_list)write_excel(metric_save_name, 'MSE', 1, MSE_list)write_excel(metric_save_name, 'PSNR', 1, PSNR_list)write_excel(metric_save_name, 'Qabf', 1, Qabf_list)write_excel(metric_save_name, 'Nabf', 1, Nabf_list)write_excel(metric_save_name, 'SSIM', 1, SSIM_list)write_excel(metric_save_name, 'MS_SSIM', 1, MS_SSIM_list)

测试一个数据集上所有对比算法的指标的程序：eval_multi_method.py

import numpy as np
from PIL import Image
from Metric import *
from natsort import natsorted
from tqdm import tqdm
import os
import statistics
import warnings
from openpyxl import Workbook, load_workbook
from openpyxl.utils import get_column_letter
warnings.filterwarnings("ignore")def write_excel(excel_name='metric.xlsx', worksheet_name='VIF', column_index=0, data=None):try:workbook = load_workbook(excel_name)except FileNotFoundError:# 文件不存在，创建新的 Workbookworkbook = Workbook()# 获取或创建一个工作表if worksheet_name in workbook.sheetnames:worksheet = workbook[worksheet_name]else:worksheet = workbook.create_sheet(title=worksheet_name)# 在指定列中插入数据column = get_column_letter(column_index + 1)for i, value in enumerate(data):cell = worksheet[column + str(i+1)]cell.value = value# 保存文件workbook.save(excel_name)def evaluation_one(ir_name, vi_name, f_name):f_img = Image.open(f_name).convert('L')ir_img = Image.open(ir_name).convert('L')vi_img = Image.open(vi_name).convert('L')f_img_int = np.array(f_img).astype(np.int32)f_img_double = np.array(f_img).astype(np.float32)ir_img_int = np.array(ir_img).astype(np.int32)ir_img_double = np.array(ir_img).astype(np.float32)vi_img_int = np.array(vi_img).astype(np.int32)vi_img_double = np.array(vi_img).astype(np.float32)EN = EN_function(f_img_int)MI = MI_function(ir_img_int, vi_img_int, f_img_int, gray_level=256)SF = SF_function(f_img_double)SD = SD_function(f_img_double)AG = AG_function(f_img_double)PSNR = PSNR_function(ir_img_double, vi_img_double, f_img_double)MSE = MSE_function(ir_img_double, vi_img_double, f_img_double)VIF = VIF_function(ir_img_double, vi_img_double, f_img_double)CC = CC_function(ir_img_double, vi_img_double, f_img_double)SCD = SCD_function(ir_img_double, vi_img_double, f_img_double)Qabf = Qabf_function(ir_img_double, vi_img_double, f_img_double)Nabf = Nabf_function(ir_img_double, vi_img_double, f_img_double)SSIM = SSIM_function(ir_img_double, vi_img_double, f_img_double)MS_SSIM = MS_SSIM_function(ir_img_double, vi_img_double, f_img_double)return EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIMif __name__ == '__main__':with_mean = Truedataroot = r'../datasets'results_root = '../Results'dataset = 'TNO'ir_dir = os.path.join(dataroot, dataset, 'ir')vi_dir = os.path.join(dataroot, dataset, 'vi')f_dir = os.path.join(results_root, dataset)save_dir = '../Metric'os.makedirs(save_dir, exist_ok=True)metric_save_name = os.path.join(save_dir, 'metric_{}.xlsx'.format(dataset))filelist = natsorted(os.listdir(ir_dir))Method_list = ['GTF', 'DIDFuse', 'RFN-Nest', 'FusionGAN', 'TarDAL', 'UMF-CMGR', 'SeAFusion', 'SwinFusion', 'U2Fusion', 'PSF']for i, Method in enumerate(Method_list):EN_list = []MI_list = []SF_list = []AG_list = []SD_list = []CC_list = []SCD_list = []VIF_list = []MSE_list = []PSNR_list = []Qabf_list = []Nabf_list = []SSIM_list = []MS_SSIM_list = []filename_list = ['']sub_f_dir = os.path.join(f_dir, Method)eval_bar = tqdm(filelist)for _, item in enumerate(eval_bar):ir_name = os.path.join(ir_dir, item)vi_name = os.path.join(vi_dir, item)f_name = os.path.join(sub_f_dir, item)print(ir_name, vi_name, f_name)EN, MI, SF, AG, SD, CC, SCD, VIF, MSE, PSNR, Qabf, Nabf, SSIM, MS_SSIM = evaluation_one(ir_name, vi_name, f_name)EN_list.append(EN)MI_list.append(MI)SF_list.append(SF)AG_list.append(AG)SD_list.append(SD)CC_list.append(CC)SCD_list.append(SCD)VIF_list.append(VIF)MSE_list.append(MSE)PSNR_list.append(PSNR)Qabf_list.append(Qabf)Nabf_list.append(Nabf)SSIM_list.append(SSIM)MS_SSIM_list.append(MS_SSIM)filename_list.append(item)eval_bar.set_description("{} | {}".format(Method, item))if with_mean:# 添加均值EN_list.append(np.mean(EN_list))MI_list.append(np.mean(MI_list))SF_list.append(np.mean(SF_list))AG_list.append(np.mean(AG_list))SD_list.append(np.mean(SD_list))CC_list.append(np.mean(CC_list))SCD_list.append(np.mean(SCD_list))VIF_list.append(np.mean(VIF_list))MSE_list.append(np.mean(MSE_list))PSNR_list.append(np.mean(PSNR_list))Qabf_list.append(np.mean(Qabf_list))Nabf_list.append(np.mean(Nabf_list))SSIM_list.append(np.mean(SSIM_list))MS_SSIM_list.append(np.mean(MS_SSIM_list))filename_list.append('mean')## 添加标准差EN_list.append(np.std(EN_list))MI_list.append(np.std(MI_list))SF_list.append(np.std(SF_list))AG_list.append(np.std(AG_list))SD_list.append(np.std(SD_list))CC_list.append(np.std(CC_list[:-1]))SCD_list.append(np.std(SCD_list))VIF_list.append(np.std(VIF_list))MSE_list.append(np.std(MSE_list))PSNR_list.append(np.std(PSNR_list))Qabf_list.append(np.std(Qabf_list))Nabf_list.append(np.std(Nabf_list))SSIM_list.append(np.std(SSIM_list))MS_SSIM_list.append(np.std(MS_SSIM_list))filename_list.append('std')## 保留三位小数EN_list = [round(x, 3) for x in EN_list]MI_list = [round(x, 3) for x in MI_list]SF_list = [round(x, 3) for x in SF_list]AG_list = [round(x, 3) for x in AG_list]SD_list = [round(x, 3) for x in SD_list]CC_list = [round(x, 3) for x in CC_list]SCD_list = [round(x, 3) for x in SCD_list]VIF_list = [round(x, 3) for x in VIF_list]MSE_list = [round(x, 3) for x in MSE_list]PSNR_list = [round(x, 3) for x in PSNR_list]Qabf_list = [round(x, 3) for x in Qabf_list]Nabf_list = [round(x, 3) for x in Nabf_list]SSIM_list = [round(x, 3) for x in SSIM_list]MS_SSIM_list = [round(x, 3) for x in MS_SSIM_list]EN_list.insert(0, '{}'.format(Method))MI_list.insert(0, '{}'.format(Method))SF_list.insert(0, '{}'.format(Method))AG_list.insert(0, '{}'.format(Method))SD_list.insert(0, '{}'.format(Method))CC_list.insert(0, '{}'.format(Method))SCD_list.insert(0, '{}'.format(Method))VIF_list.insert(0, '{}'.format(Method))MSE_list.insert(0, '{}'.format(Method))PSNR_list.insert(0, '{}'.format(Method))Qabf_list.insert(0, '{}'.format(Method))Nabf_list.insert(0, '{}'.format(Method))SSIM_list.insert(0, '{}'.format(Method))MS_SSIM_list.insert(0, '{}'.format(Method))if i == 0:write_excel(metric_save_name, 'EN', 0, filename_list)write_excel(metric_save_name, "MI", 0, filename_list)write_excel(metric_save_name, "SF", 0, filename_list)write_excel(metric_save_name, "AG", 0, filename_list)write_excel(metric_save_name, "SD", 0, filename_list)write_excel(metric_save_name, "CC", 0, filename_list)write_excel(metric_save_name, "SCD", 0, filename_list)write_excel(metric_save_name, "VIF", 0, filename_list)write_excel(metric_save_name, "MSE", 0, filename_list)write_excel(metric_save_name, "PSNR", 0, filename_list)write_excel(metric_save_name, "Qabf", 0, filename_list)write_excel(metric_save_name, "Nabf", 0, filename_list)write_excel(metric_save_name, "SSIM", 0, filename_list)write_excel(metric_save_name, "MS_SSIM", 0, filename_list)write_excel(metric_save_name, 'EN', i + 1, EN_list)write_excel(metric_save_name, 'MI', i + 1, MI_list)write_excel(metric_save_name, 'SF', i + 1, SF_list)write_excel(metric_save_name, 'AG', i + 1, AG_list)write_excel(metric_save_name, 'SD', i + 1, SD_list)write_excel(metric_save_name, 'CC', i + 1, CC_list)write_excel(metric_save_name, 'SCD', i + 1, SCD_list)write_excel(metric_save_name, 'VIF', i + 1, VIF_list)write_excel(metric_save_name, 'MSE', i + 1, MSE_list)write_excel(metric_save_name, 'PSNR', i + 1, PSNR_list)write_excel(metric_save_name, 'Qabf', i + 1, Qabf_list)write_excel(metric_save_name, 'Nabf', i + 1, Nabf_list)write_excel(metric_save_name, 'SSIM', i + 1, SSIM_list)write_excel(metric_save_name, 'MS_SSIM', i + 1, MS_SSIM_list)

在上述三个程序中均需调用 Metric.py函数：

import numpy as np
from scipy.signal import convolve2d
from Qabf import get_Qabf
from Nabf import get_Nabf
import math
from ssim import ssim, ms_ssimdef EN_function(image_array):# 计算图像的直方图histogram, bins = np.histogram(image_array, bins=256, range=(0, 255))# 将直方图归一化histogram = histogram / float(np.sum(histogram))# 计算熵entropy = -np.sum(histogram * np.log2(histogram + 1e-7))return entropydef SF_function(image):image_array = np.array(image)RF = np.diff(image_array, axis=0)RF1 = np.sqrt(np.mean(np.mean(RF ** 2)))CF = np.diff(image_array, axis=1)CF1 = np.sqrt(np.mean(np.mean(CF ** 2)))SF = np.sqrt(RF1 ** 2 + CF1 ** 2)return SFdef SD_function(image_array):m, n = image_array.shapeu = np.mean(image_array)SD = np.sqrt(np.sum(np.sum((image_array - u) ** 2)) / (m * n))return SDdef PSNR_function(A, B, F):A = A / 255.0B = B / 255.0F = F / 255.0m, n = F.shapeMSE_AF = np.sum(np.sum((F - A)**2))/(m*n)MSE_BF = np.sum(np.sum((F - B)**2))/(m*n)MSE = 0.5 * MSE_AF + 0.5 * MSE_BFPSNR = 20 * np.log10(255/np.sqrt(MSE))return PSNRdef MSE_function(A, B, F):A = A / 255.0B = B / 255.0F = F / 255.0m, n = F.shapeMSE_AF = np.sum(np.sum((F - A)**2))/(m*n)MSE_BF = np.sum(np.sum((F - B)**2))/(m*n)MSE = 0.5 * MSE_AF + 0.5 * MSE_BFreturn MSEdef fspecial_gaussian(shape, sigma):"""2D gaussian mask - should give the same result as MATLAB's fspecial('gaussian',...)"""m, n = [(ss-1.)/2. for ss in shape]y, x = np.ogrid[-m:m+1, -n:n+1]h = np.exp(-(x*x + y*y) / (2.*sigma*sigma))h[h < np.finfo(h.dtype).eps*h.max()] = 0sumh = h.sum()if sumh != 0:h /= sumhreturn hdef vifp_mscale(ref, dist):sigma_nsq = 2num = 0den = 0for scale in range(1, 5):N = 2**(4-scale+1)+1win = fspecial_gaussian((N, N), N/5)if scale > 1:ref = convolve2d(ref, win, mode='valid')dist = convolve2d(dist, win, mode='valid')ref = ref[::2, ::2]dist = dist[::2, ::2]mu1 = convolve2d(ref, win, mode='valid')mu2 = convolve2d(dist, win, mode='valid')mu1_sq = mu1*mu1mu2_sq = mu2*mu2mu1_mu2 = mu1*mu2sigma1_sq = convolve2d(ref*ref, win, mode='valid') - mu1_sqsigma2_sq = convolve2d(dist*dist, win, mode='valid') - mu2_sqsigma12 = convolve2d(ref*dist, win, mode='valid') - mu1_mu2sigma1_sq[sigma1_sq<0] = 0sigma2_sq[sigma2_sq<0] = 0g = sigma12 / (sigma1_sq + 1e-10)sv_sq = sigma2_sq - g*sigma12g[sigma1_sq<1e-10] = 0sv_sq[sigma1_sq<1e-10] = sigma2_sq[sigma1_sq<1e-10]sigma1_sq[sigma1_sq<1e-10] = 0g[sigma2_sq<1e-10] = 0sv_sq[sigma2_sq<1e-10] = 0sv_sq[g<0] = sigma2_sq[g<0]g[g<0] = 0sv_sq[sv_sq<=1e-10] = 1e-10num += np.sum(np.log10(1+g**2 * sigma1_sq/(sv_sq+sigma_nsq)))den += np.sum(np.log10(1+sigma1_sq/sigma_nsq))vifp = num/denreturn vifpdef VIF_function(A, B, F):VIF = vifp_mscale(A, F) + vifp_mscale(B, F)return VIFdef CC_function(A,B,F):rAF = np.sum((A - np.mean(A)) * (F - np.mean(F))) / np.sqrt(np.sum((A - np.mean(A)) ** 2) * np.sum((F - np.mean(F)) ** 2))rBF = np.sum((B - np.mean(B)) * (F - np.mean(F))) / np.sqrt(np.sum((B - np.mean(B)) ** 2) * np.sum((F - np.mean(F)) ** 2))CC = np.mean([rAF, rBF])return CCdef corr2(a, b):a = a - np.mean(a)b = b - np.mean(b)r = np.sum(a * b) / np.sqrt(np.sum(a * a) * np.sum(b * b))return rdef SCD_function(A, B, F):r = corr2(F - B, A) + corr2(F - A, B)return rdef Qabf_function(A, B, F):return get_Qabf(A, B, F)def Nabf_function(A, B, F):return Nabf_function(A, B, F)def Hab(im1, im2, gray_level):hang, lie = im1.shapecount = hang * lieN = gray_levelh = np.zeros((N, N))for i in range(hang):for j in range(lie):h[im1[i, j], im2[i, j]] = h[im1[i, j], im2[i, j]] + 1h = h / np.sum(h)im1_marg = np.sum(h, axis=0)im2_marg = np.sum(h, axis=1)H_x = 0H_y = 0for i in range(N):if (im1_marg[i] != 0):H_x = H_x + im1_marg[i] * math.log2(im1_marg[i])for i in range(N):if (im2_marg[i] != 0):H_x = H_x + im2_marg[i] * math.log2(im2_marg[i])H_xy = 0for i in range(N):for j in range(N):if (h[i, j] != 0):H_xy = H_xy + h[i, j] * math.log2(h[i, j])MI = H_xy - H_x - H_yreturn MIdef MI_function(A, B, F, gray_level=256):MIA = Hab(A, F, gray_level)MIB = Hab(B, F, gray_level)MI_results = MIA + MIBreturn MI_resultsdef AG_function(image):width = image.shape[1]width = width - 1height = image.shape[0]height = height - 1tmp = 0.0[grady, gradx] = np.gradient(image)s = np.sqrt((np.square(gradx) + np.square(grady)) / 2)AG = np.sum(np.sum(s)) / (width * height)return AGdef SSIM_function(A, B, F):ssim_A = ssim(A, F)ssim_B = ssim(B, F)SSIM = 1 * ssim_A + 1 * ssim_Breturn SSIM.item()def MS_SSIM_function(A, B, F):ssim_A = ms_ssim(A, F)ssim_B = ms_ssim(B, F)MS_SSIM = 1 * ssim_A + 1 * ssim_Breturn MS_SSIM.item()def Nabf_function(A, B, F):Nabf = get_Nabf(A, B, F)return Nabf

完整demo下载地址：https://download.csdn.net/download/fovever_/87530213

指标复现不易，如果您使用了我们的程序请点赞，收藏，关注三连哦。

如果您觉得我们的程序对您的科研有用，请移步Github:https://github.com/Linfeng-Tang 引用博主的相关文献，整理不易，您的支持是我们持续开源相关程序的动力。

如有疑问可添加博主QQ：2458707789@qq.com 请备注：（ 学校+姓名 ），否则不予通过好友申请。