SL0分类稀疏表示的图像修复算法  

Classified Sparse Representation of Image Inpainting by SL0 Algorithm

作　　者：屠雅丽 唐向宏[1] 张东[1] 蔡倩[1] 任玉升[2] 

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018 [2]浙江水利水电学院电气工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2016年第1期32-36,56,共6页Journal of Hangzhou Dianzi University：Natural Sciences

摘　　要：将SL0算法应用于图像修复.在传统的SL0算法基础上,采用近似双曲正切函数去近似l0范数,利用共轭梯度法求解该函数,实现高精度的重构,提出一种利用SL0算法的分类稀疏表示图像修复算法.算法首先将图像分块,并根据图像块的梯度信息和局部方差依次对图像块进行分类;然后分别对每一类图像块样本用K-SVD字典训练得到相对应的过完备字典;最后利用改进的SL0算法进行图像修复.仿真实验结果表明,算法对图像修复后的质量有较大的提高,更符合人眼视觉效应.This paper proposes a novel image inpainting algorithm by SL0 algorithm. Based on traditional SL0 algorithm, it takes use of approximate hyperbolic tangent function to approach l0 norm and uses conjugate gradient to solve this function, which are helpful to improve the reconstructive accuracy. The proposed algorithm firstly divides the damaged image into blocks and classifies them in turn through the gradient information and local variance of image block; then trains the each type image block samples with K-SVD to get the corresponding over complete dictionary; finally, utilizes the modified SL0 algorithm to image inpainting. The experiment results show that the proposed algorithm in image inpainting is greatly improved the image quality and the images are more in accordance with visual effect.

关 键 词：SL0 K-SVD 梯度 局部方差 分类训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]