A local texture-constrained super-resolution method

This paper proposes a local texture constrained super-resolution method for the reconstruction of high-resolution image. Through the learned low/high-resolution patches from training images, the intended high resolution patches are reconstructed using neighbor embedding method. The major contributions of this paper are: 1) Local Binary Pattern (LBP) is adopted to classify the patches into different categories, only those patches who have the same pattern with the input patches are used as candidates; 2) Structural SIMilarity (SSIM) metric which can find the patches with texture most similar to the input is used to search the k most suitable patches in the corresponding category. Experiments show that LBP index can provide proper candidate patches and SSIM metric is better than other metric in finding the most texture similarity patches.