Cognitive And Neural Computing Modeling Of Image Storage And Retrieval Inspired By The Mechanism Of Human Brain Memory

With the development of neuroscience and brain science,scientists are now on the threshold of a better understanding of how the human memory works.Modeling human memory mechanism for information storage and retrieval has become a hot research issue,and various of methods have been put forward.However,from the perspective of cognitive neuroscience,there are few researches about visual image information storage and retrieval based on human memory mechanism.Thus,this thesis mainly focuses on cognitive and neural computing modeling of image storage and retrieval inspired by the mechanism of human brain memory.The main contributions of this thesis are as follows:1.First,the mechanism of human brain memory is summarized.The roles of hippocampus and the medial temporal lobe cortex in the brain for human memory mechanism are analyzed.Then,several memory modeling methods are studied.Finally,a cognitive neural computing framework of information storage and retrieval based on CLS model is proposed.2.A feature representation model base on sparse coding is proposed.The visual perception mechanism is modeled using sparse coding algorithm,and a fast dictionary learning algorithm is applied to solve the problem of sparse coding.Meanwhile,two methods of reducing features based on randomly features selection and features selection by density are proposed for tackling the issue of large calculation.The proposed model significantly accelerates the solutions to dictionary learning and sparse coding.3.A cognitive and neural computing model of information storage and retrieval based on SOINN(Self-Organizing Incremental Neural Network)is proposed.SOINN is able to incrementally learn new knowledge without destroying the learned knowledge,and there is no need to predefine the structure or size of the network,thus helping to imitate the information storage and retrieval process of the human brain.Experimental results on the benchmark data set demonstrate the effectiveness of the proposed method.