Algorithm Of Image Retrieval On Hadoop

How to retrieve user’s interesting multimedia data from mass data efficiently is aresearch hot spot now. For image retrieval, traditional method is base on keywords ordescriptions, in fact, it is base on text retrieval and its accuracy is not satisfied. Content-basedimage retrieval algorithm has started to research ten years ago, Google has such applicationnow, but the retrieval accuracy and speed needs to improve. Locality Sensitive HashingAlgorithm (LSH) is one of the content-based image retrieval algorithm, it is also one of thepopular approximate nearest neighbor query algorithm. LSH is base on image’s feature and dowell in retrieve high-dimension data.Hadoop cloud platform is a popular cloud platform currently, which mainly consists oftwo parts. One is Distributed File System (Hadoop Distributed File System, HDFS), the otheris the parallel computing model: MapReduce framework. HDFS distributed store data inmany nodes on the cluster, so as to provide a high-capacity sotrage. MapReduce framework,divide the task into many sub-tasks while each sub-task is caculated by different node,through this parallel computing model to improve the computational efficiency.The main subject of research is image retrieval algorithms base on Hadoop. Image dataset distributed stored in Hadoop platform. By creating image index files to retrieve imagerapidly. This paper presents LSH optimization algorithm based on image’s global features(GIST features), LSH algorithm based on image’s Scale-invariant features(SIFT features) andinverted index files algorithm based on image’s “Bay of Words”(BoW)model, thenimplemented these three algorithms on Hadoop platform via using MapReduce parallelcomputing framework.In this paper, through a large number of experiments, analysis and comparison of thesethree algorithms’ precision，recall ratio and performance. Experimental results show that theabove three retrival algorithm implement on Hadoop platform can respond quickly and havegood results. LSH algorithm based on SIFT features has the highest precision, the lowestrecall ratio and the longest run time. LSH algorithm based on GIST features has higherprecision, the highest recall ratio and the shortest run time. Inverted index files algorithmbased on BoW model has the lowest precision, higher recall ratio and shorter run time.