콘텐츠 본문
논문 국내 국내전문학술지(KCI급) Enhancing Image Compression with Foveal Vision: A Multi-Focus FPSNR Assessment and Attention-Based Neural Network
- 학술지 구분 국내전문학술지(KCI급)
- 게재년월 2024-12
- 저자명 Andri Agustav Wirabudi, Haechul Choi
- 발행처명 한국방송미디어공학회
- 발행국가 국내
- 논문언어 외국어
- 전체저자수 2
논문 초록 (Abstract)
In the field of image and video compression, the objective is to achieve a balance between compression efficiency and the quality of reconstructed images.
The commonly used quality assessment method in this field is the Peak Signal-to-Noise Ratio (PSNR), which, however, has a limitation in that it only considers the differences in pixel values.
To address this, our research introduces the Foveal Peak Signal-to-Noise Ratio (F_PSNR), a visual perception-based approach that reflects human foveal vision.
Specifically, we propose a multi-focus F_PSNR assessment method that incorporates the visual characteristics of humans for images containing multiple objects of interest.
Additionally, we suggest a model that integrates an attention mechanism focusing on the quality of objects of interest into the existing neural network-based compression method to enhance perception-based quality.
Experimental results using the KODAK dataset demonstrate that applying the attention mechanism to existing methods can enhance the human-perceptual compression efficiency of neural networks.
