Iteration forecast orthogonal matching pursuit algorithm

To improve orthogonal matching pursuit algorithm, the mainstream direction is to select several atoms which is matching the residual in each iteration.However, there is few in-depth research about selection criteria of atoms.Generally, it is to select K atoms in atom library which have largest correlation coefficients with residual, or select all atoms whose correlation coefficients with residual are greater than a certain threshold. Based on orthogonal matching pursuit algorithm,by statistical method, this paper researches the relationship between the two atoms of two adjacent iterations, and concludes that their correlation coefficients with the earlier residual have interval feature, which has a guiding significance on choosing more atoms in one iteration. And it can support the prediction of the atom in the next iteration with high probability. Based on this, this paper presents the iteration forecast orthogonal matching pursuit algorithm. Experimental results show that, compared with other matching pursuit algorithm, its time-consuming is greatly reduced while the reconstruction accuracy has not significantly reduced.