Saliency guided bit allocation for image compression


An image has a foreground and a background region. In image compression standards like JPEG, JPEG2000 it is possible to adaptively allocate coding bits. For example, one may use more bits per pixel in the foreground as compared with pixels in the background. This may potentially increase the perceived image quality, even when the total bitrate for an image is kept the same. So-called saliency maps generalize the binary foreground/background to continuous level and may serve as a more appropriate guide for allocating bits, e.g., per JPEG 8x8 image block. The aim of this project is to design different strategies for this approach and to evaluate the effectiveness by practical user studies.


  • Obtain data base (images, saliency maps)
  • For a set of fixed bitrates do: compute JPEG/JPEG2000 codes with different bit allocations based on saliency
  • Lab and/or crowd sourcing studies using pairwise stimuli
  • Quantify the importance of the salient region (repeat analysis of the paper of Alers et al)
  • Optimize the bit allocation (transform the analysis of Li et al from video to images)

Perspectives for Bachelor/Master theses:

For the master it is required to expand the research by adapting the bit allocation also to different distortions like blurriness, low contrast, poor color, noise.


Solid practical knowledge and experience with algorithms. Programming in MATLAB, C, and/or C++.


  1. Li, Zhicheng, Qin, Shiyin and Itti, Laurent, "Visual attention guided bit allocation in video compression,"Image and Vision Computing, 29(1), pp.1-14, 2011.
  2. Alers, Hani, Redi, Judith, Liu, Hantao and Heynderickx, Ingrid, "Studying the effect of optimizing image quality in salient regions at the expense of background content," Journal of Electronic Imaging, 22(4), pp.1-10, 2013.