An Improved Training Sample Set Construction Method
Li Shen, Zhai Jiaojiao · Journal of Physics Conference Series · 2019
Sample set construction is an important step in the super-resolution reconstruction algorithm based on dictionary learning, which has an important impact on the training of dictionary and the effect of image reconstruction. When the sample set was built, the similarity between the sample blocks was not taken into account, which resulted in the redundancy of the sample set, thus increasing the time overhead of the follow-up dictionary training. To solve this problem, this paper proposes an effective method for constructing sample sets. By setting a reasonable threshold value of Euclidean distance similarity, the proposed method can ensure that the constructed sample set have structural anisotropy and diversity. When the number of pre-set blocks is the same, the training time of dictionary is reduced to about 50% of the original method, and the quality of image reconstruction is improved.