navalmartin_mir_vision_utils

A collection of various utilities for working with images in the mir project. The provided utilities use to a large extent the python PIL library.

Acknowledgements

The project incorporates the following repositories

• image-quality https://github.com/ocampor/image-quality (for BRISQUE)
• imutils: https://github.com/PyImageSearch/imutils (for various utilities with OpenCV)

Dependencies

The general dependencies are:

• numpy
• Pillow
• scipy
• scikit-image
• libsvm

In addition, utilities for working to PyTorch and OpenCV exists but you need to install these dependencies yourself. The mir_vision_config provides various configuration flags to customize the API. These are

• WITH_TORCH
• WITH_CV2
• WITH_SKIMAGE_VERSION

Installation

Installing the utilities via pip

pip install navalmartin-mir-vision-utils

For a specific version use

pip install navalmartin-mir-vision-utils==x.x.x

You can uninstall the project via

pip3 uninstall navalmartin-mir-vision-utils

How to use

Below are some use-case samples. You can find more in the examples.

Using image_utils

Various image utilities are provided in this module. The example below showcases some.

from pathlib import Path
from navalmartin_mir_vision_utils import (is_valid_pil_image_file,
                                          get_pil_image_size,
                                          get_img_files,
                                          pil_image_to_bytes_string,
                                          create_thumbnail_from_pil_image)

from navalmartin_mir_vision_utils.mir_vison_io import get_md5_checksum

if __name__ == '__main__':

    image_file = Path("/home/alex/qi3/mir-engine/datasets/cracks_v_3_id_8/train/cracked/img_9_9.jpg")
    image = is_valid_pil_image_file(image=image_file)
    if image is not None:
        print("The provided image is OK")
        image_size = get_pil_image_size(image=image)
        print(f"Image size is {image_size}")
    else:
        print("The provided image is NOT OK")

    base_path = Path("/home/alex/qi3/mir-engine/datasets/cracks_v_3_id_8/train/cracked/")
    image_files = get_img_files(base_path=base_path)
    print(f"There are {len(image_files)} in {base_path}")

    # calculate file checksum
    image_checksum = get_md5_checksum(file=image_file)
    print(f"Calculated MD5 checksum {image_checksum}")

    image_checksum = get_md5_checksum(file=image.tobytes())
    print(f"Calculated MD5 checksum {image_checksum}")
    
    # create a thumbnail
    image = create_thumbnail_from_pil_image(max_size=(50, 50),
                                            image_filename=image_file)

    image.show()

Using image_transformers

from pathlib import Path

from navalmartin_mir_vision_utils import (plot_pil_image,
                                          plot_pytorch_image,
                                          is_valid_pil_image_from_bytes_string,
                                          load_img,
                                          ImageLoadersEnumType,
                                          pil_image_to_bytes_string,
                                          pil_to_torch_tensor,
                                          pils_to_torch_tensor)
from navalmartin_mir_vision_utils.mir_vision_config import WITH_TORCH


if __name__ == '__main__':

    image_path = Path("../../tests/test_data/img_18_3.jpg")
    image = load_img(path=image_path, loader=ImageLoadersEnumType.PIL)

    print("Showing image from normal loading...")
    plot_pil_image(image=image, title="Image from file")

    image_bytes = pil_image_to_bytes_string(image=image)
    image = is_valid_pil_image_from_bytes_string(image_byte_string=image_bytes)

    print("Showing image from bytes...")
    plot_pil_image(image=image, title="Image from byte string")

    if WITH_TORCH:
        # convert the image to a torch tensor
        torch_tensor = pil_to_torch_tensor(image=image)

        print(f"Torch tensor size={torch_tensor.size()}")
        plot_pytorch_image(image=torch_tensor, title="PyTorch tensor image")

        # load another image
        image_path = Path("../../tests/test_data/img_2_6.jpg")
        image_2 = load_img(path=image_path, loader=ImageLoadersEnumType.PIL)
        plot_pil_image(image=image_2, title="Second image")

        # make the two images torch tensors
        torch_tensors = pils_to_torch_tensor(images=[image, image_2])
        print(f"Torch tensor size={torch_tensors.size()}")

Compute basic image statistics

from pathlib import Path
from navalmartin_mir_vision_utils import load_img, ImageLoadersEnumType
from navalmartin_mir_vision_utils.statistics import compute_image_statistics, fit_gaussian_distribution_on_image

if __name__ == '__main__':

    image_path = Path("/home/alex/qi3/mir-engine/datasets/cracks_v_3_id_8/train/cracked/img_9_9.jpg")
    image = load_img(path=image_path, loader=ImageLoadersEnumType.PIL)

    print(f"Image size {image.size}")
    print(f"Image bands {image.getbands()}")

    image_stats = compute_image_statistics(image)
    print(f"Image channel mean {image_stats.mean}")
    print(f"Image channel var {image_stats.var}")
    print(f"Image channel median {image_stats.median}")

    channels_fit = fit_gaussian_distribution_on_image(image=image)
    print(f"Gaussian distribution channel fit: {channels_fit}")

Compute image quality calculation

Currently, only the BRISQUE algorithm is supported. The implementation from https://github.com/ocampor/image-quality has been integrated into the utilities.

from pathlib import Path

from navalmartin_mir_vision_utils import load_img, ImageLoadersEnumType
from navalmartin_mir_vision_utils.image_quality import brisque

if __name__ == '__main__':
    image_path = Path("/home/alex/qi3/mir-engine/datasets/cracks_v_3_id_8/train/cracked/img_9_9.jpg")
    image = load_img(path=image_path, loader=ImageLoadersEnumType.PIL)
    brisque_score = brisque.score(image)
    print(brisque_score)

Issues

• rescale() got an unexpected keyword argument 'multichannel'

This issue may be related to the version of the skimage package you have installed. You can check with version is installed on your system by using

import skimage
print(skimage.__version__)

or use the variable WITH_SKIMAGE_VERSION in mir_vision_config.