自然图像稀疏编码模型研究综述-《郑州大学学报(工学版)》

文章信息/Info

Author(s):: ZOU Bai-xian1; MIAO Jun2; 1.College of Arts & Science，Beijing Union University，Beijing 10083，China; 2.Key Laboratory of Intelligent InformationProcessing，Institute of Computing ’Technology ,Chinese Academy of Seiences ,Beijing 100190，China

摘要:: 根据建模出发点的不同,把各种建模方法分为模拟视觉系统模型、统计分析模型两大类方法,根据不同的目标、不同的模型特征和结构,把模拟视觉系统的稀疏编码模型分又为最大似然概率、目标函数优化、Gabor小波基函数、超完备基、神经网络、分层稀疏编码六类模型.根据模型学习方法的不同,统计分析模型又分为独立分析、非负矩阵分解以及特定特征的稀疏编码3种模型.针对上述各种模型进行了介绍、分析和研究,并归纳总结不同方法的主要特点,最后进行了展望.

Abstract:: Tdifferent modeling objectives,characteristics,the sparse coding models for simulation of the visual system are classified into the following sixcategories ,those are maximum likelihood models,the optimization models of the objective functions,the func-tion model of Gabor wavelet basis,super-complete basis models,the models of neural networks,and the levelsparse coding model. In accordance with learning methods for models, the statistical analysis models are divid-ed into independent analysis models,non-negative matrix factorization models,as well as specific characteris-tics of sparse coding models. All the various models are introduced,analyzed and studied,the main featuresof different models are Summarized and compared,and their future prospects are proposed.