Skip to content
Snippets Groups Projects
utils.py 3.32 KiB
Newer Older
Matthieu Muller's avatar
Matthieu Muller committed
"""A file containing some useful functions for the project.
This file should NOT be modified.
"""


from skimage.metrics import peak_signal_noise_ratio, structural_similarity
from skimage.io import imread, imsave
import numpy as np

Matthieu Muller's avatar
Matthieu Muller committed
from src.checks import check_data_range, check_rgb, check_shape, check_path, check_png
Matthieu Muller's avatar
Matthieu Muller committed


def normalise_image(img: np.ndarray) -> np.ndarray:
    """Normalise the values of img in the interval [0, 1].

    Args:
        img (np.ndarray): Image to normalise.

    Returns:
        np.ndarray: Normalised image.
    """
    return (img - np.min(img)) / np.ptp(img)


def load_image(file_path: str) -> np.ndarray:
    """Loads the image located in file_path.

    Args:
        file_path (str): Path of the file containing the image. Must end by '.png'.

    Returns:
        np.ndarray: The loaded image.
    """
    check_path(file_path)
    check_png(file_path)

    return normalise_image(imread(file_path))


def save_image(file_path: str, img: np.ndarray) -> None:
    """Saves the image located in file_path.

    Args:
        file_path (str): Path of the file in which the image will be saved. Must end by '.png'.
        img (np.ndarray): Image to save.
    """
    check_path(file_path.split('/')[-2])
    check_png(file_path)

    imsave(file_path, (img * 255).astype(np.uint8))


def psnr(img1: np.ndarray, img2: np.ndarray) -> float:
    """Computes the PSNR between img1 and img2 after some sanity checks.
    img1 and img2 must:
        - have the same shape;
        - be in range [0, 1].

    Args:
        img1 (np.ndarray): First image.
        img2 (np.ndarray): Second image.

    Returns:
        float: PSNR between img1 and img2.
    """
    check_shape(img1, img2)
    check_data_range(img1)

    return peak_signal_noise_ratio(img1, img2, data_range=1)


def ssim(img1: np.ndarray, img2: np.ndarray) -> float:
    """Computes the SSIM between img1 and img2 after some sanity checks.
    img1 and img2 must:
        - have the same shape;
        - be in range [0, 1];
        - be 3 dimensional array with 3 channels.

    Args:
        img1 (np.ndarray): First image.
        img2 (np.ndarray): Second image.

    Returns:
        float: SSIM between img1 and img2.
    """
    check_shape(img1, img2)
    check_data_range(img1)
    check_rgb(img1)

    return structural_similarity(img1, img2, data_range=1, channel_axis=2)


####
####
####

####      ####                ####        #############
####      ######              ####      ##################
####      ########            ####      ####################
####      ##########          ####      ####        ########
####      ############        ####      ####            ####
####      ####  ########      ####      ####            ####
####      ####    ########    ####      ####            ####
####      ####      ########  ####      ####            ####
####      ####  ##    ######  ####      ####          ######
####      ####  ####      ##  ####      ####    ############
####      ####  ######        ####      ####    ##########
####      ####  ##########    ####      ####    ########
####      ####      ########  ####      ####
####      ####        ############      ####
####      ####          ##########      ####
####      ####            ########      ####
####      ####              ######      ####

# 2023
# Authors: Mauro Dalla Mura and Matthieu Muller