image super -resolution as sparse representation of raw image patches代码解读之字典训练

1、

clear all; clc; close all;
addpath(genpath('RegularizedSC'));

TR_IMG_PATH = 'Data/Training';

dict_size   = 512;          % dictionary size
lambda      = 0.15;         % sparsity regularization？？？？？？？？怎么用
patch_size  = 5;            % image patch size
nSmp        = 100000;       % number of patches to sample
upscale     = 2;            % upscaling factor上采样

% randomly sample image patches
[Xh, Xl] = rnd_smp_patch(TR_IMG_PATH, '*.bmp', patch_size, nSmp, upscale);

% prune减少 patches with small variances,临界值 threshould chosen based on the
% training data
[Xh, Xl] = patch_pruning(Xh, Xl, 10);

% joint sparse coding 
[Dh, Dl] = train_coupled_dict(Xh, Xl, dict_size, lambda, upscale);
dict_path = ['Dictionary/D