Image super-resolution based on Gabor-transformed-neighbor-embedding
WU Xiao-mi · 2014
An algorithm based on Gabor transformation is proposed to improve the performance of image super-resolution(SR).In this algorithm,the local features of images can be obtained by the convolutions of the image patches and the multi-scale and multi-orientation Gabor filters.The K-nearest-neighbors(KNNs)can be found out for each test image patch by computing the Euclidean distances between it and image patches in the training set.The relationships between the low-resolution(LR)and the highresolution(HR)image patch pairs can be obtained by joint learning method.The required HR image patch can be computed by the linear combination of the corresponding KNNs.Experimental results show that the Gabor features which extracted from image patch can help the test image patch to find its more accurate KNNs and thus improve the performance of SR.