CVMMIVApr 20, 2022

Multi-Scale Features and Parallel Transformers Based Image Quality Assessment

arXiv:2204.09779v16 citationsh-index: 24Has Code
Originality Synthesis-oriented
AI Analysis

This work addresses the challenge of assessing image quality for multimedia applications, representing an incremental improvement by combining existing techniques.

The paper tackles the problem of image quality assessment by integrating multi-scale feature extraction with transformer networks, and demonstrates that the proposed method outperforms existing algorithms on datasets including PIPAL.

With the increase in multimedia content, the type of distortions associated with multimedia is also increasing. This problem of image quality assessment is expanded well in the PIPAL dataset, which is still an open problem to solve for researchers. Although, recently proposed transformers networks have already been used in the literature for image quality assessment. At the same time, we notice that multi-scale feature extraction has proven to be a promising approach for image quality assessment. However, the way transformer networks are used for image quality assessment until now lacks these properties of multi-scale feature extraction. We utilized this fact in our approach and proposed a new architecture by integrating these two promising quality assessment techniques of images. Our experimentation on various datasets, including the PIPAL dataset, demonstrates that the proposed integration technique outperforms existing algorithms. The source code of the proposed algorithm is available online: https://github.com/KomalPal9610/IQA

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes