Probabilistic Principal Component Analysis and Channel Attention for End-to-End Image Compression Optimization
In recent years, deep learning has shown significant progress for image compression compared to traditional image compression methods. Although conventional standard-based methods are still used, they are limited in handling repetitive patterns and complex calculations, which can lead to image recon...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11072174/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In recent years, deep learning has shown significant progress for image compression compared to traditional image compression methods. Although conventional standard-based methods are still used, they are limited in handling repetitive patterns and complex calculations, which can lead to image reconstruction issues. In this study, we propose a novel learning-based image compression method that integrates both channel attention (CA) and probabilistic principal component analysis (PPCA) blocks as core components to enhance encoding efficiency. PPCA is used to focus on essential features and manage noise. Unlike traditional PCA, PPCA’s probabilistic approach better preserves meaningful data structure, enhancing compression and robustness. The CA mechanism in our model emphasizes significant image features by prioritizing dominant pixel values, allowing the compression process to retain essential details while minimizing less relevant information. Furthermore, a foveated image quality assessment metric is proposed, prioritizing visually significant regions to enhance the evaluation of dominant information guided by attention mechanisms and to assess the impact of the CA and PPCA blocks on image reconstruction. Experimental results demonstrate that the proposed method obtained significant coding efficiency across various metrics on the Kodak and Tecnick datasets compared with state-of-the-art methods. |
|---|---|
| ISSN: | 2169-3536 |