Single image super resolution (SR) based on sparse representation is a promising technique where the SR problem is solved by searching for the most robust representation of a signal in terms of atoms in a dictionary. However, first and second-order derivatives are always used as features for patches to be trained as dictionaries and super-resolved patches are reconstructed using dictionaries and sparse representation of these features. In this paper, a novel single image SR algorithm based on sparse representation with considering the effect of significant features is proposed. Therefore, high frequency details are preserved using discrete wavelet transform (DWT). In addition, an intermediate process is also proposed to learn finer dictionaries and thereby estimate the sharper and more detailed super-resolved image. The dictionaries are constructed from the distinctive features using K-SVD dictionary learning algorithm. The intermediate process uses approximation subband. Thus, constructed dictionaries contain so much more significant information and the interpolated high frequency components are corrected. Therefore, the intermediate process restores the high frequency details better in super-resolved images. The proposed algorithm was tested on 'Set14' dataset. Owing to DWT, the proposed algorithm recovers the edges better as well as improving the computational efficiency. The quantitative and visual results show the superiority and competitiveness of the proposed method over the simplest techniques and state-of-art SR algorithms. Experimental time comparisons with the state-of-art algorithms validate the effectiveness of the proposed approach.