Intensity-Sensitive Similarity Indexes for Image Quality Assessment

Xiaotong Li, Wesley Armour2022 · DOI · arXiv

Abstract

The importance of Image quality assessment (IQA) is ever increasing due to the fast paced advances in imaging technology and computer vision. Among the numerous IQA methods, Structural SIMilarity (SSIM) index and its variants are better matched to the perceived quality of the human visual system. However, SSIM methods are insufficiently sensitive, when images contain low information, where the important information only occupies a low proportion of the image while most of the image is noise-like, which is common in scientific data. Therefore, we propose two new IQA methods, InTensity Weighted SSIM index and Low-Information Similarity Index, for such low information images. In addition, auxiliary indexes are proposed to assist with the assessment. The application of these new IQA methods to natural images and field-specific images, such as radio astronomical images, medical images, and remote sensing images, are also demonstrated. The results show that our IQA methods perform better than state-of-the-art SSIM methods for differences in high-intensity parts of the input images and have similar performance to that of the original and gradient-based SSIM for differences in low-intensity parts.

Introduction: The Challenge of Sparse Scientific Data

Standard Image Quality Assessment (IQA) tools, most notably the Structural Similarity Index (SSIM), were developed to align with the Human Visual System (HVS). These metrics excel at evaluating natural photographs by prioritizing features humans find salient and ignoring imperceptible noise. However, scientific data—ranging from radio astronomy to medical imaging—presents a different challenge.

In these fields, images are often "low-information," meaning the scientifically relevant signal (e.g., a distant star or a lung lesion) occupies only a small fraction of the image area. The remainder is typically a noise-like background. Because standard SSIM averages information across the entire image, the sheer volume of background noise "dilutes" the signal. This leads to high similarity scores even when the critical data points differ significantly, a phenomenon known as the "noise dilution effect."

Low-information scientific images where signal is sparse against a noisy background.

Proposed Methodology: ITW-SSIM and LISI

To overcome the limitations of HVS-based metrics, we propose two new indexes designed to prioritize high-intensity features:

1. InTensity Weighted SSIM (ITW-SSIM)

ITW-SSIM modifies the traditional SSIM framework by introducing intensity-based weighting functions. By applying Gaussian, tanh, or sigmoid functions, the metric assigns significantly higher importance to pixels with high intensity during the calculation of mean, variance, and covariance. This ensures that the "bright" parts of the image—where the signal resides—dominate the final assessment.

Weighting functions (Gaussian, tanh, and sigmoid) used in ITW-SSIM to prioritize high-intensity pixels.

2. Low-Information Similarity Index (LISI)

LISI represents a more fundamental shift. It is structured as a ratio where the numerator captures "effective similarity" (the match between high-intensity pixels) and the denominator represents the total "amount of important information." This design ensures the index is near zero when comparing noise and approaches one only when high-intensity features align perfectly.

3. Auxiliary Indexes

We also introduce a Sensitivity Index ($sensi$) to quantify the relative sensitivity of new metrics against baselines, and a Direction Index ($direc$) to track whether overall intensity is increasing or decreasing—a vital feature for monitoring physical processes like tumor growth or soil desiccation.

Experimental Results and Applications

We evaluated these metrics across diverse datasets, including natural images (TID2013), radio astronomy (GLEAM), medical scans (Lung-PET-CT-Dx), and remote sensing (QuickDRI).

  • Radio Astronomy: In tests comparing astronomical images with varying noise levels, standard SSIM yielded a misleadingly high similarity of 0.9707. In contrast, LISI provided a score of 0.0309, accurately reflecting that the informative signal components were distinct.
  • Remote Sensing: Using the QuickDRI dataset to monitor drought in California, ITW-SSIM (using sigmoid/tanh weighting) detected drying trends at least one week earlier than standard SSIM. This increased sensitivity could provide critical lead time for wildfire prevention and water management.
  • Medical Imaging: The metrics proved effective in tracking changes in CT scans over time, with the Direction Index providing clear indicators of lesion progression or regression.

Conclusion

The transition from HVS-based weighting to intensity-sensitive metrics allows for a more objective assessment of scientific data integrity. While ITW-SSIM and LISI are not intended to replace SSIM for general photography, they offer a powerful new toolkit for researchers in fields where signal sparsity and background noise are inherent challenges. Future work will explore integrating these metrics into iterative imaging algorithms to optimize signal extraction.

Summary updated

Abstract

The importance of Image quality assessment (IQA) is ever increasing due to the fast paced advances in imaging technology and computer vision. Among the numerous IQA methods, Structural SIMilarity (SSIM) index and its variants are better matched to the perceived quality of the human visual system. However, SSIM methods are insufficiently sensitive, when images contain low information, where the important information only occupies a low proportion of the image while most of the image is noise-like, which is common in scientific data. Therefore, we propose two new IQA methods, InTensity Weighted SSIM index and Low-Information Similarity Index, for such low information images. In addition, auxiliary indexes are proposed to assist with the assessment. The application of these new IQA methods to natural images and field-specific images, such as radio astronomical images, medical images, and remote sensing images, are also demonstrated. The results show that our IQA methods perform better than state-of-the-art SSIM methods for differences in high-intensity parts of the input images and have similar performance to that of the original and gradient-based SSIM for differences in low-intensity parts. Different similarity indexes are suitable for different applications, which we demonstrate in our results.
BibTeX
@inproceedings{li2022intensity,
  author = {Xiaotong Li and Wesley Armour},
  title = {{Intensity-Sensitive Similarity Indexes for Image Quality Assessment}},
  booktitle = {2022},
  year = {2022},
  doi = {10.1109/icpr56361.2022.9956093},
}
Figure