Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-12 Cooperative journals: 《计算机应用研究》
Abstract: Aiming at the problem of poor quality and contrived details in super-resolution reconstruction by external learning, this paper proposes a single image super-resolution based on compressed sensing and similarity constraint. Firstly, a classified dictionary based on measurement domain is proposed by using the linear relationship between the measurement domain and the frequency domain in compressed sensing, which improves the representation of dictionaries, then, the non-local similarity is used in the reconstruction process, the sparsity under the feature dictionary and similar block information are combined as priori information to regularize super-resolution reconstruction, and finally the high resolution image is recovered. Experimental results show that the reconstructed image has rich details and defined edges, and the subjective quality is good.