| 摘 要: 由于图像数据的爆炸式增长,使得图像压缩算法得到蓬勃发展与广泛应用。系统综述了自2018年以来基于深度学习的图像压缩算法研究工作;首先,对比传统算法,阐述了深度学习算法的优势;其次,基于图像压缩、算法经典框架,从编解码器优化和熵模型优化两方面对国内外研究方法进行对比和论。研究发现,编解码器能优化提升算法的特征提取与图像重建能力,熵模型优化则致力于提升压缩效率和降低算法复杂度。最后,分析了图像压缩算法面临的挑战及未来的研究方向。 |
| 关键词: 图像压缩 深度学习 超先验架构 |
|
中图分类号: TP391
文献标识码: A
|
|
| Survey of Deep Learning-Based Image Compression Algorithms |
|
WANG Hailong, MA Lingkun, ZHANG Weichuan
|
(School of Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, China)
whl19829165517@163.com; malingkun@sust.edu.cn; zwc2003@163.com
|
| Abstract: Driven by the exponential growth of image data, image compression algorithms have experienced rapid development and widespread application. This paper systematically reviews research on deep learning-based image compression algorithms since 2018. First, by comparison with traditional compression schemes, the advantages of deep learning- based methods are elaborated. Second, based on the classic framework of image compression algorithms,domestic and international research methods are compared and discussed from two perspectives: encode-r decoder optimization and entropy model optimization. It is revealed that encode-r decoder optimization enhances the algorithm’s feature extraction and image reconstruction capabilities, while entropy model optimization focuses on improving compression efficiency and reducing algorithmic complexity. Finally, the challenges and future research directions of
image compression algorithms are analyzed. |
| Keywords: image compression deep learning hyperprior architecture |
