一种基于压缩感知的图像去马赛克算法

Compressed Sensing Based Image Demasaicing Algorithm

文中提出了一种基于分块压缩感知的图像去马赛克算法。该算法首先将Bayer色彩滤镜阵列采样值等效为压缩感知理论中感知矩阵采样所得的压缩数据。其次通过结合Bayer色彩滤镜阵列自身特点训练分块稀疏字典。最后在训练所得稀疏字典的基础上利用分块压缩感知重构算法便可精确重构出Bayer色彩滤镜阵列采样结果。由于训练所得稀疏字典能有效降低与Bayer色彩滤镜采样阵列之间的相关性,因此文中所提出的图像去马赛克算法能有效对单一Bayer色彩滤镜阵列采样值进行重构。通过实验验证表明:新的图像去马赛克算法明显优于传统插值算法,重构所得彩色图像在去除假色影响的同时能完整保留原始图像的细节信息。

A new block compressed sensing （BCS） based demasaicing algorithm is proposed for reconstructing the sampling data of Bayer color filter array （Bayer CFA）.In the new demasaicing algorithm,we firstly take the sampling data of Bayer CFA as the compressed measurement one in the BCS,and then,utilize the character of the Bayer CFA to train a sparse dictionary.Finally,the sampling data of Bayer CFA is reconstructed through using the trained sparse dictionary and BCS reconstruction method.Since there is a low coherence between the Bayer CFA and the sparse dictionary,so we are able to effectively eliminate masaicing in the reconstructed image.Experiments show that the new demasaicing algorithm has a better performance than the interpolation algorithms.The reconstructed image using the new demasaicing algorithm can eliminate bias effect and keep the detail color information of the original image.