utils.py

import numpy as np
import os
from scipy.stats import multivariate_normal
from matplotlib import pyplot as plt
from PIL import Image

from skimage import io as skio 
from skimage import color as skco
from skimage import transform

import warnings 

import ipdb

##################################################################################
#
#  image read/write operation
#
##################################################################################
def img_route(img_id = "001", method = "LR", scale = 2):
    # dataset_route = "/home/e/Eulring/GitProject/Mixture-Function-Sparse-Representation/dataset/Set14"
    # dataset_route = "/Users/eulring/Efile/Dataset/Set14"
    dataset_route = "./Set14"
    
    img_route = os.path.join(dataset_route, "image_SRF_"+str(scale))
    img_name = '/img_' + img_id + '_SRF_' + str(scale) + '_' + method + '.png'
    print(img_route + img_name)
    return img_route + img_name



#=================================================================================
#  turn the img_route to the normalized numpy array of y channel directly 
def numpy_from_img_route(img_route, imagecut = False, batch_size = (6, 6)):
    img_rgb = Image.open(img_route)
    img_ycbcr = rgb2ycbcr(img_rgb)
    img1 = np.array(img_ycbcr)
    y = img1[:, :, 0]
    
    if imagecut == True:
        y = imgcut(y, batch_size = batch_size)
    
    if np.max(y) > 1: y = y / 255.0
        
    return y











##################################################################################
#
#  Evalution
#
##################################################################################

#=================================================================================
# RMSE
def RMSE(img1, img2):
    if len(img1.shape) == 3:
        img1_y = skco.rgb2ycbcr(img1)[:, :, 0]
    else: 
        img1_y = img1
    if len(img2.shape) == 3:
        img2_y = skco.rgb2ycbcr(img2)[:, :, 0]
    else: 
        img2_y = img2

    img_dif = img1_y - img2_y
    
    rmse = np.sqrt(np.mean(img_dif**2))
    return rmse

 
def PSNR(img1, img2):
    rmse = RMSE(img1, img2)
    return 20 * np.log10(255.0/rmse)

def quick_eval(img_O, img_id = '001', scale = 2):
    # ipdb.set_trace()
    img_B = skio.imread(img_route(img_id = img_id, method='bicubic', scale = scale))
    img_H = skio.imread(img_route(img_id = img_id, method='HR', scale = scale))
    
    img_H, img_O = equal_size(img_H, img_O)
    img_H, img_B = equal_size(img_H, img_B)
    # ipdb.set_trace()
    PSNR_1 = PSNR(img_H, img_O)
    PSNR_2 = PSNR(img_H, img_B)
    
    SSIM_1 = SSIM(img_H, img_O)
    SSIM_2 = SSIM(img_H, img_B)
    
    print('PSNR/SSIM for Bicubic Interpolation: %f dB', PSNR_2, SSIM_2);
    print('PSNR/SSIM for Sparse Representation Recovery: %f dB', PSNR_1, SSIM_1);

    
#=================================================================================
# PSNR
# refer : https://github.com/aizvorski/video-quality/blob/master/psnr.py
def PSNR1(img1, img2):
    import math
    mse = np.mean( (img1 - img2) ** 2 )
    if mse == 0:
        return 100
    PIXEL_MAX = 255.0
    return 20 * math.log10(PIXEL_MAX / math.sqrt(mse))

def PSNR0(img1, img2):
    from skimage.measure import compare_psnr
    return compare_psnr(img1, img2, 255)

def PSNR2(img1, img2):
    diff = img1 - img2
    mse = np.mean(np.square(diff))
    psnr = 10.0 * np.log10(255.0 * 255 / mse)
    return psnr



#=================================================================================
# SSIM
# refer : https://github.com/aizvorski/video-quality/blob/master/ssim.py
def SSIM(img1, img2, data_range = 1.0, multichannel=False):
    img1_ = img1.copy()
    img2_ = img2.copy()
    if np.max(img1_) < 2.0 : img1_*=255.0
    if np.max(img2_) < 2.0 : img2_*=255.0
    data_range = 255.0
    if len(img1_.shape) == 3 : multichannel=True
    import skimage
    
    return skimage.measure.compare_ssim(img1_, img2_, data_range=data_range, multichannel=True)
'''
    return compare_ssim(
        img1, img2,
        win_size=11,
        gaussian_weights=True,
        multichannel=True,
        data_range=1.0,
        K1=0.01,
        K2=0.03,
        sigma=1.5)
'''
##################################################################################
#
#  Other
#
##################################################################################


def equal_size(img1, img2):
    w1, h1 = img1.shape[0], img1.shape[1]
    w2, h2 = img2.shape[0], img2.shape[1]
    
    w = min(w1, w2)
    h = min(h1, h2)
    
    if len(img1.shape) == 3:
        img1 = img1[0 : w, 0 : h, :]
        img2 = img2[0 : w, 0 : h, :]
    else:
        img1 = img1[0 : w, 0 : h]
        img2 = img2[0 : w, 0 : h]
    
    return img1, img2