基于峭度准则的稀疏编码算法与仿真研究  

ON KURTOSIS CRITERION-BASED SPARSE CODING ALGORITHM AND ITS SIMULATION

作　　者：段均[1] 

机构地区：[1]南充职业技术学院信息与管理工程系,四川南充637100

出　　处：《计算机应用与软件》2015年第9期324-327,333,共5页Computer Applications and Software

摘　　要：针对自然图像压缩收敛速度慢的问题,提出一种新的基于峭度的绝对值和固定系数方差的稀疏编码SC(Sparse Coding)算法。该算法采用稀疏性惩罚函数来表示峭度大小,同时保证了图像特征系数的分散性与独立性,并维持图像重构误差和稀疏惩罚函数之间的平衡,能够更有效地提取图像的边缘特征和局部特征。通过选取合适的特征基函数,有利于加快所提出的SC网络的收敛速度。应用该算法可以成功地提取自然图像的特征基向量,进一步利用特征系数的稀疏性,有效实现自然图像的压缩。仿真实验结果表明,与基于标准独立分量分析(ICA)和离散余弦变换(DCT)的图像压缩方法相比,基于峭度准则的稀疏编码图像压缩方法具有较快的收敛速度及较好的有效性和实用性。As the natural image compression has the problem of slow convergence rate, we put forward a new sparse coding （SC） algorithm which is based on the absolute value of kurtosis and fixed parameter variance. This algorithm adopts sparse penalty function to indicate the kurtosis size, and meanwhile ensures the dispersibility and independence of the image feature coefficients, as well as maintains the balance between image reconstruction error and sparse penalty function, thus is able to extract the edge features and local features of the image more efficiently. By selecting proper feature base function, it is conducive to expedite the convergence speed of the proposed SC network. To apply this algorithm can successfully extract the feature base vector of nature image, as well as further utilises the sparseness of the feature coefficient and effectively achieves the compression of nature image. Simulation experiment results showed that, compared with the image compression methods based on standard isolated component analysis （ICA） and discrete cosine transform （DCT）, the sparse coding image compression method based on kurtosis criterion had faster convergence rate and better effectiveness and practicability.

关 键 词：稀疏编码 峭度准则 特征提取 图像压缩 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]