Compressed Sensing Image Recovery using Dictionary Learning

Chengfu Huo, Rong Zhang, Yin Dong, and Anzhou Hu


Compressed Sensing, Recovery Algorithm, Reference Image, Dictionary


The compressed sensing theory states that a sparse signal can be recovered accurately from a small number of linear measurements. With the assumption of having a relative image for reference, this paper presents an improved recovery algorithm for images sampled by compressed sensing. First, a coarse image is recovered from the measurements by employing an initial DCT dictionary. Secondly, the DCT dictionary is updated based on the reference image. The two steps are operated alternately to enhance the dictionary sparsify ability, and so as to improve the recovery performance. Experimental results on remote sensing images have demonstrated the effectiveness of this proposal. More importantly, it can be found from this paper that, even though the updated dictionary is coherent, a better recovery performance can still be achieved due to its superior sparsify ability.

Important Links:

Go Back