Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Compressed image quality assessment using stacking

S. Farhad Hosseini-Benvidi, Hossein Motamednia, Azadeh Mansouri, Mohammadreza Raei, Ahmad Mahmoudi-Aznaveh

TL;DR

The results of the Full-Reference and No-Reference models are aggregated to improve the proposed Full-Reference method for compressed image quality evaluation and the accuracy of the quality benchmark of the clic2024 perceptual image challenge was achieved, which illustrates the effectiveness of the proposed fusion-based approach.

Abstract

It is well-known that there is no universal metric for image quality evaluation. In this case, distortion-specific metrics can be more reliable. The artifact imposed by image compression can be considered as a combination of various distortions. Depending on the image context, this combination can be different. As a result, Generalization can be regarded as the major challenge in compressed image quality assessment. In this approach, stacking is employed to provide a reliable method. Both semantic and low-level information are employed in the presented IQA to predict the human visual system. Moreover, the results of the Full-Reference (FR) and No-Reference (NR) models are aggregated to improve the proposed Full-Reference method for compressed image quality evaluation. The accuracy of the quality benchmark of the clic2024 perceptual image challenge was achieved 79.6\%, which illustrates the effectiveness of the proposed fusion-based approach.

Compressed image quality assessment using stacking

TL;DR

The results of the Full-Reference and No-Reference models are aggregated to improve the proposed Full-Reference method for compressed image quality evaluation and the accuracy of the quality benchmark of the clic2024 perceptual image challenge was achieved, which illustrates the effectiveness of the proposed fusion-based approach.

Abstract

It is well-known that there is no universal metric for image quality evaluation. In this case, distortion-specific metrics can be more reliable. The artifact imposed by image compression can be considered as a combination of various distortions. Depending on the image context, this combination can be different. As a result, Generalization can be regarded as the major challenge in compressed image quality assessment. In this approach, stacking is employed to provide a reliable method. Both semantic and low-level information are employed in the presented IQA to predict the human visual system. Moreover, the results of the Full-Reference (FR) and No-Reference (NR) models are aggregated to improve the proposed Full-Reference method for compressed image quality evaluation. The accuracy of the quality benchmark of the clic2024 perceptual image challenge was achieved 79.6\%, which illustrates the effectiveness of the proposed fusion-based approach.
Paper Structure (4 sections, 4 figures, 1 table)

This paper contains 4 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Conclusion
  4. References

Figures (4)

  • Figure 1: An example of a situation in the validation dataset where the conventional methods, namely PSNR, LPIPS, and Tres, selected the second image as the one more similar to the reference image. However, our method aligns more closely with human judgment and correctly identifies the first image as the better match.
  • Figure 2: On the left side of the image, among the four methods chosen, three methods (NIQE, TOPIQ, and PieAPP) considered the second image superior in quality. However, our proposed method correctly identified the higher quality image with the assistance of a single vote from the hyperIQA method. The right side of the image presents a similar situation, but this time the TOPIQ method's vote was considered. This demonstrates that our method is not biased towards a specific method.
  • Figure 3: The supporter methods are applied just on supported False results
  • Figure 4: We conducted a thorough investigation on 15 quality assessment methods. By analyzing all the possible combinations of these 15 methods and representing the highest accuracy achieved by subsets of a fixed size as a scatter plot, we observed that the accuracy reaches a saturation when using subsets consisting of 4 methods. Additionally, we found that increasing the number of features does not lead to a significant improvement in accuracy.