Design And Implementation Of Image Retrieval System Based On Hadoop

With the continuous popularization of the Internet and the extensive application of social media,the media data on the Internet has increased by injecting well,and the massive information has been generated.Therefore,in the massive image,retrieving the required image has become a major problem.At present,image retrieval has become a research hotspot.The traditional content-based image retrieval methods generally use a single server,and takes a long time,which has long been unable to meet the development needs of the people.Therefore,the research direction began to turn to distributed processing,because the distributed processing can quickly carry on image processing and reduce the time spent on image processing calculation.In this paper,it uses the convolution neural network model to extract the eigenvectors of the image,then use the local sensitive hash algorithm to calculate the hash codes corresponding to the vectors.Then,the Hamming distance between different hash codes is used to judge the similarity distance between the images,and the candidate image set is retrieved.In order to further improve the accuracy,the feature vector of the candidate image set,not the hash code,is used of similarity measure.Finally,the adjacent images of the query image are retrieved.In order to ensure retrieval efficiency and scalability of the retrieval system,this paper adopts the parallel computing ability of Hadoop cloud platform and implements the distributed computing of image retrieval algorithm by using MapReduce to solve the computing limit in single environment.The Hadoop cluster is set up to test the algorithm designed,and compare with the running time of Hadoop and single machine environment.The precision and recall of the image retrieval algorithm are analyzed.The experimental results show that when using the image retrieval algorithm designed in this paper,the retrieval rate is higher.Hadoop cluster can be used to run the image retrieval algorithm flexibly according to the user's needs.The system designed in this paper is scalability.Compared with the single-machine environment,the running time of the Hadoop system is greatly shortened on the dataset,and with the use of the sample image,the more obvious the advantages of performance of the Hadoop system's distributed processing.And the efficiency of the algorithm has been significantly improved.With the increase of nodes' number,the execution time of the algorithm is further shortened,which shows the better expansibility of the system.