将稀疏自动编码器用于置换混叠图像盲分离  

Blind Separation for Permuted Alias Image with the Sparse Auto-encoders

作　　者：段新涛[1,2] 李飞飞[1] 刘尚旺[1] 彭涛[1] 王婧娟[1] 

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453007 [2]河南省高校"计算智能与数据挖掘"工程技术研究中心,河南新乡453007

出　　处：《信号处理》2016年第5期608-617,共10页Journal of Signal Processing

基　　金：国家自然科学基金资助项目(U1204606;U1304607)资助;河南省教育厅高等学校重点科研项目(15A520080)资助;河南省科技攻关项目(142102210565)资助;河南师范大学博士科研启动基金(qd12138)资助

摘　　要：针对一类置换区域含噪声的置换混叠图像,提出一种基于稀疏自动编码器的算法来自动检测和分离含噪声的置换区域。对含噪声的置换混叠图像进行分块,获取输入数据集。构建稀疏自动编码器网络,通过数据集训练参数,获得解码后的置换混叠图像。将解码后图像与原置换混叠图像作差运算得到差图像,通过检测差图像来确定置换区域,采用自适应阈值化操作分离出含噪声的置换区域,实现对置换区域的自动检测和分离。实验结果表明,采用本算法在置换区域位置、大小、个数和所含噪声类型、大小均未知的情况下,能有效地分离出含噪声的置换区域。Focused on the issue that a class of permuted alias image infecting noise in permuting region,an algorithm about permuted alias image blind separation based Sparse Auto-encoder was proposed to detect and separate permuting region with noise automatically. Firstly,permuted alias image was divided into blocks,and the input dataset was obtained. Then the decoded permuted alias image was obtained by constructing Sparse Auto-encoder network model and training parameters using the dataset. The permuting region is found out by detecting the subtraction image,which is defined as difference between the decoded image and original permuted alias image. The permuting region was separated from the permuted alias image automatically by adaptive threshold operating. Experimental results show that the permuting region can be separated from the permuted alias image efficiently,not affected by location,size,number of permutation region,noise type,noise level on permuting region.

关 键 词：盲分离 置换混叠图像 稀疏自动编码器 深度学习 噪声检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]