Image Semantic Hierarchical Classification And Retrieval Based On Formal Concept Analysis

With the popularization of Internet technology and intelligent camera electronic devices,the image information exposed in human life is expanding.In order to effectively manage image resources,this paper proposes a classification and retrieval system for multi-semantic complex images.These include four main research problems:multi-semantic image annotation based on discriminated model,multi-semantic image annotation based on semantic propagation,multi-semantic hierarchical classification and multi-semantic image retrieval.Starting from the idea of discriminated model,this paper proposes an image annotation algorithm based on semantic migration and fusion neural network.Image multi-scale semantic fusion neural network and semantic migration neural network based on semantic linear fitting are designed in the algorithm.In the semantic fusion network,this paper designs a pattern mining algorithm based on frequent concept lattices.It can cluster the multi-scale image blocks with the same semantics and obtain the fusion features,which solves the semantic multi-scale problem in the discriminated model.The semantic migration network is inspired by human cognition.It can learn the unknown semantic training discriminated model from the known semantics to solve the semantic limitation problem.In the direction of image annotation based on semantic propagation,this paper designs an image annotation algorithm based on weighted semantic neighbor set from the two research purposes of ensuring the accuracy and integrity of semantic propagation.The image annotation algorithm proposed in this paper constructs a weighted semantic neighbor set by dividing the semantic importance of the training set,which ensures the comprehensiveness of the annotation semantics.Finally,by designing the semantic word frequency adaptive threshold function,the accuracy of passing semantic tags is guaranteed.In this paper,a multi-semantic hierarchical classification algorithm based on partial order structure is designed,to combine it with the semantic information obtained byimage annotation.In the multi-semantic hierarchical classification algorithm,images not only achieve hierarchical classification according to semantic commonality,but also the association rules between semantics are considered,and the hierarchical sharing of semantic information is realized.Therefore,the hierarchical classification structure designed in this paper not only has a simple and clear hierarchical relationship,but also can reflect the subordination of semantics in the data set.For complex multi-semantic images,this paper designs a two-dimensional multi-semantic image retrieval algorithm from two aspects: semantic information and visual information.The two-dimensional retrieval matrix proposed in the algorithm can consider the semantic similarity and visual similarity of the image in parallel.Combining the unique retrieval method designed in this paper with the two-dimensional retrieval matrix can ensure that the multi-semantic images with the highest ranking can balance visual similarity and semantic similarity.