Research On Image Retrieval Algorithm Based On Discrete Graph Hashing

With the rapid development of the Internet,the scale of image data is exploding.It is of great significance to study how to efficiently retrieve interesting images for users in massive image data.The traditional content-based image retrieval algorithm mainly uses the content features of the image to match the similarity,which faces to many problems like the feature semantic gap,high feature dimension,the large storage space,and the low retrieval efficiency.Using the Hashing-based image retrieval algorithm can effectively compensate for the above shortcomings.In order to solve indexing accuracy and efficien-cy,this paper proposes an image retrieval algorithm based on discrete graph hashing.The main research results are as follows:Aiming at the problem of feature extraction in image retrieval,a multi-view non-negative feature fusion algorithm based on Laplacian graph is proposed.Non-negative matrix factorization technique is adopted to transform each view feature,and the non-negative feature expression is more powerful.Each view feature is only a representation of image semantic information.The Laplacian model is used to embed the non-negative features of a variety of views into a unified latent space in which the merged features can express image semantic features better.In the construction of Laplacian model,the“an-chor" technique is introduced to reduce the computational complexity of Laplace matrix.According to the model,the experiments on open data set are carried out to verify that the multi-feature fusion algorithm studied in this paper can achieve higher accuracy than single-feature retrieval algorithm.In order to construct efficient index,a discrete graph hashing image retrieval algorith-m with supervised machine learning is proposed.By studying the hash functions,the data features in the original space are mapped to the Hamming space to preserve the similarity of the data.The similarity of hash code is calculated in the Hamming space.Hashing func-tion is learned by using the supervised machine learning method and the discrete graph optimization framework.Then,use discrete objective function to avoid the traditional method,which uses the label information of the data to express the semantic information of the image.Using the "relaxation" strategy leads to a lower quality of hash code and improves the precision of retrieval.The hash code of all training samples is generated bit by bit through using discrete cycle descent method.Then the generating speed of the hash code is improved.The experimental results show that the proposed algorithm has high efficiency in image retrieval.In order to verify the feature validity of multi-view non-negative feature fusion,a discrete graph hashing algorithm is used to retrieve the feature.Experimental results show that this feature can be combined with the discrete graph hashing algorithm to improve the retrieval accuracy in image retrieval.