COPEX-high-rate-compression-quality-metrics 0.0.4

COPEX high rate compression quality metrics

COPEX High Rate Compression Quality Metrics

This package provides quality metrics for high rate compression.

Installation

pip install COPEX_high_rate_compression_quality_metrics

Library usage

Comparison of Two Multiband TIFF Images via LRSP (L:LPIPS, R:RMSE, S:SSIM, P:PSNR)

the following shows concrete example of use of the library

Steps:

1. Import necessary libraries
2. Initialize the model and specify the file paths
3. Define preprocessing functions
4. Load the TIFF files
5. Calculate the metrics

1 Library install / import

use "pip install COPEX-high-rate-compression-quality-metrics" to install the library

# Array handler
import numpy as np
# to have a json formated output
import json
import math
# File handler
from skimage import io

# File path handler
import os

# Metrics and utils
import COPEX_high_rate_compression_quality_metrics.metrics as COPEX_metrics
import COPEX_high_rate_compression_quality_metrics.utils as COPEX_utils

2 LPIPS initialization

# LPIPS initialization
loss_fn = COPEX_metrics.initialize_LPIPS()

3 File path definition

# Specify file paths here
file_path1 = os.path.join('T28PGV_20160318T111102_B04_20m.tif')
file_path2 = os.path.join('T28PGV_20160318T111102_B04_20m_ter.tif')

4 File loading

#load images and show shapes
image1 = io.imread(file_path1)
print(file_path1," [shape =",image1.shape,", min =",np.min(image1), ", max =",np.max(image1), ", dtype = ",image1.dtype,"]")
image2 = io.imread(file_path2)
print(file_path2," [shape =",image2.shape,", min =",np.min(image2), ", max =",np.max(image2), ", dtype = ",image2.dtype,"]")


# checking if images have the same shape
if image1.shape != image2.shape:
    raise ValueError("Les deux images doivent avoir les mêmes dimensions.")
print("images loaded with success.")

File visualization (optional)

COPEX_utils.display_multiband_tiffs(image1, image2)

5 metrics calculation

# Calculate all metrics
lpips_values,lpips_value = COPEX_metrics.calculate_lpips_multiband(image1, image2,loss_fn)
mean_ssim = COPEX_metrics.calculate_ssim_multiband(image1, image2)
psnr_value = COPEX_metrics.calculate_psnr(image1, image2)
rmse_value = COPEX_metrics.calculate_rmse(image1, image2)

Results interpretation

see VT-P382-SLD-003-E-01-00_COPEX_DCC_PM3_20230630.pdf for more informations about metrics weeknesses

LPIPS : (identical images) 0 <==========> 1 (completely different images) lower is better [very good LPIPS do not mean that images are not totaly different pixel wise]

RMSE : (identical images) 0 <==========> +inf (completely different images) lower is better [different kind of degradations can give the same score, do not capture blurring]

SSIM : (completely different images) -1 <==========> 1 (identical images) higher is better [sensible to little local distorions, sensible to noise differences]

PSNR : (completely different images) 0 <==========> +inf (identical images) higher is better [sensible to Big local differences]

data = {
    "files paths":{
        "file1":file_path1,
        "file2":file_path2
        },
    "metrics":{
        "LPIPS":lpips_value,
        "RMSE":rmse_value,
        "SSIM":mean_ssim,
        "PSNR":str(psnr_value) if math.isinf(psnr_value) else psnr_value     
    }
}
json_data = json.dumps(data, indent=4)

print(json_data)

    {
        "files paths": {
            "file1": "T28PGV_20160318T111102_B04_20m.tif",
            "file2": "T28PGV_20160318T111102_B04_20m_ter.tif"
        },
        "metrics": {
            "LPIPS": 0.038381848484277725,
            "RMSE": 192.01308995822703,
            "SSIM": 0.9817911582274915,
            "PSNR": 26.491058156522357
        }
    }