A Survey on Learning End-to-end Lossy Image Compression

PhD Qualifying Examination


Title: "A Survey on Learning End-to-end Lossy Image Compression"

by

Mr. Ka Leong CHENG


Abstract:

Lossy image compression has been a popular and fundamental technique for 
computer vision and image processing since the digital information era. Many 
traditional image compression codec (e.g., JPEG, JPEG2000, and BPG) are widely 
and commonly used in practical applications. Recently, researchers tend to 
develop deep-learning based algorithms for lossy image compression due to their 
superiority of compression rates compared to the classical ones. Some methods 
focus on how to build a better encoding or decoding transformation between the 
source images and the latent codes; some pay special attention to develop a 
better entropy model to estimate the latent code distribution more effectively 
and accurately.

This survey presents a comprehensive review of learned lossy image compression 
methods. We first give the historical background of lossy image compression and 
formulate the image compression problem using the fundamental rate-distortion 
theory in data compression. Next, we introduce several important contributions 
in recent research progresses. We then summarize the current state-of-the-art 
approaches. After that, we further mention some common training and evaluation 
strategies plus a popular evaluation platform for image compression. This 
survey ends with a final conclusion.


Date:  			Friday, 22 July 2022

Time:                  	10:00am - 12:00noon

Zoom Meeting: 
https://hkust.zoom.us/j/93227315905?pwd=RG9ZaGE1TXJGUThtaUlpVXk2dmZvdz09

Committee Members:	Dr. Qifeng Chen (Supervisor)
 			Dr. Hao Chen (Chairperson)
 			Prof. Raymond Wong
 			Dr. Dan Xu


**** ALL are Welcome ****