Dictionary based encoding of cosmological images

The pioneering theory of Compressed Sensing (CS) provides a framework for ill-posed inverse problems and allows for the recovery of sparse signals from a set of measurements. Its applicability to astronomy datasets was recognised from its infancy. In this work, CS techniques are used to aid in the construction of an optimised dictionary that is capable of encoding cosmological images. A learning algorithm that automatically determines and adapts the size of the repository according to the provided training set, is presented. Use of the robust and fast StOMP ℓ1 minimization method is made for the recovery of sparse one dimensional signals. The results suggest that accurate reconstructions with very low residual errors can be obtained.