Improved Image Retrieval Algorithm Based On HVLAD

With the rapid development of computer technology,the multimedia technology is also flourishing.Multimedia devices play an increasingly important role in people's lives.Vision is the most important means to access information.Therefore,all these information is closely related to all aspects of life,and provides great convenience for people's working and learning.However,Technology development has also brought more challenges for people,Now huge amount of digital image information has been produced everyday,how to classify very well in such a large-scale image database,how to target an image in a short time has become the hot research topic for scholars.Therefore,image retrieval technology has been developed rapidly,and has become a hot research field in computer technology.Since the text of the image retrieval technology to turned to based image retrieval technology,content-based image retrieval has a good effect compared with text based image retrieval technology to reduce the problem of man-made complex labor and subjective error,has been the mainstream technology in the field of image retrieval.In the field of content-based image retrieval,image feature expression algorithm determines the efficiency of image retrieval,which is commonly used in Bag of Words(BoVW)and Vector of Locally Aggregated Descriptors(VLAD).Visual word bag model in image retrieval application has a good effect,so it has been widely used,but with the improvement of local visual features,more cluster centers are needed in the application,it will be a sharp increasement in the amount of calculation,simutaneously by focusing on the feature quantity information and makes a lot of feature information loss,weakening the ability of expression.Local vector descriptor(Vector of Locally Aggregated Descriptors,VLAD)is the expression of the global image information,combined clustering through local visual features and its distribution to the distance value,compared to the visual word bag model has a higher accuracy.But because of the restriction of the number of visual words and the lack of global descriptor distribution of the information status,it also affects the efficiency of image retrieval.In order to solve the above problems,this paper combines the HVLAD algorithm with the data distribution entropy to improve the expression of image features.HVLAD can effectively solve the problem of missing feature information in VLAD due to small visual codebook and insufficient feature space division.Data distribution entropy is obtained by computing the coordinates and scale 3D information in each clustering feature space,and the entropy of data distribution contains information about the distribution of global feature points.The combination of the above two algorithms can further guarantee more image information in image expression.In this paper,the K-means clustering algorithm of SIFT feature images in the database,obtaining the visual words codebook,classifying visual feature space,and then K-means operation in each feature space,visual word HVLAD codebook clustering,clustering HVLAD feature space.In each HVLAD feature space,the sum of the distance between the local feature point and the cluster center is calculated,and the image expression of the sub cluster is obtained.On the basis of HVLAD,the entropy of data distribution is added,and a better image expression is obtained by combining the HVLAD with the entropy.In this paper,a Holiday database and Oxford database in the field of image retrieval are applied to verify the as a representative database experiment.The experimental results show that compared with the original VLAD and HVLAD,it has a better retrieval effect by adding entropy.Therefore,the algorithm has a better retrieval efficiency and a better user experience after the principle analysis and experimental verification in this paper.