A shrinkage learning approach for single image super-resolution with overcomplete representations
ECCV'10 Proceedings of the 11th European conference on Computer vision: Part II
| Hi-index | 0.00 |
It is well-known that wavelet transforms provide sparse decompositions overmany types of image regions but not over image singularities/edges that manifest themselves along curves. It is now widely accepted that, on 2D piecewisesmooth signals, wavelet compression performance is dominated by coefficientsover edges. Research in this area has focused on two tracks, each suffering fromissues related to translation invariance. Methods that directly model high order coefficient dependencies over edges have to combat aliasing issues, and newtransforms that have been designed lose their full strength if they are not usedin a translation invariant fashion. In this paper we combine these approachesand use translation invariant, overcomplete representations to predict waveletedge coefficients. By starting with the lowest frequency band of an l level waveletdecomposition, we reliably estimate missing higher frequency coefficients overpiecewise smooth signals. Unlike existing techniques, our approach does notmodel edges directly but implicitly obtains boundaries by aggressively determining regions where the utilized translation invariant decomposition is sparse.