3D Model Retrieval System Based On Relevance Feedback And Clustering Analysis Technology: Research And Implement

Follow the internet and the information technology change with eachpassingday,peoplehave manyways toretrieveinformationfrom theinternet.But people's demands of requiring information begin to change higher andhigher. For example the accuracy of the information and timeliness andefficiency of information.As the retrieval pattern on pictures or text ininternet,people are not interesting it any more.Next generation retrievalengine has been attached importance to bypeople more and more.The mediastream of sound, image and video is make the leading actor ofinformation.Beacuse 3D games, virtual reality,movie enterprises, pharmacyand machine manufactures have abroad apply the computer technology,thenthe 3D Models are gradually make the forth media data type.In thetremendousamountofwebinformation,itisverydifficultyhowtosearchthesuited 3D models.People are very concerned about this problem.And howquicklyandaccuracysearchthemodelsbecamethemainproblemthatpeopleare concerned about.In this background,we study the fashion 3D retrievalengine and we put forward the imagine of 3D model retrieval system basedonrelevancefeedbackandclusteringanalysistechnology.Thisdissertationcontraposestwoproblemsof3Dmodelretrievalenginethat people are concened about.The first problem is the retrievalefficiency.The second problem is or not importing the be concerned withpeople,that is the relevance feedback technology,so these technologies makeheighten the retrieval effect.When the model database begin to sharpincrease,if we don not use some mechanism to heighten the retrievalefficiency,the retrieval system must have been collapsed.And it is essentialthat we use the relevance feedback technology in the field of the 3D modelretrieval engine,because we have study the CRIB,and we find that relevancefeedback technology is abroad applied.So we study the relevance feedbackand clustering techniques base on above problems.And we design the 3DmodelretrievalsystemwithJ2EEandModel2pattern.The core idea of the relevance feedback technology is that import theconcened with people in the retrieval process. This technique try to providethe good ways what will resolve the contact that is between low-layer featureof models and high-layer semantemes of models. In the retrievalprocess ,through the alternations of the computer with people,system obtainthe semantic information of users,and change the search with the mark ofusers'feedback.This will make the search results transfer to users.The mainway is that users are added after the first search,and uses give the marks offeedback that make the system bingen to study,and bengin to changeclever.But fashion 3D model engines are designed with content-basedretrieval,system can automatic manage the feature extraction and retrieve. Sothere are big gap between the retrieval results. Though relevance feedbackhas been a lively topic of research in text retrieval and in image retrieval, ithas hardly been explored in 3D model retrieval.We are aware of only a fewalgorithms that specifically target 3D models.We try to realize our relevancefeedbackalgorithmswithlineardiscriminantanalysisorSVM.Thealgorithm oflineardiscriminant analysis is someapproaches belongto the family of query point movement methods,for which the task of thelearnerconsists infinding,at everyround,abatter querypoint totogetherwitha re-weightingof the individual dimensions of the description space.And thismethod use positive training examples it becomes a classificationproblem.SVM is another assume that is an elegant way to deal withnon-linerly is to be use reproducing kernel based algorithms.A kernel basedone-class SVM as density estimator for positive examples was shown in tooutperformthewhiteningtransformbasedlinear/quadraticmethod.Clustering analysis technology is a method when there are not anytraining examples system begin to make classification for the modeldatabases.And users retrieve models in the clustering results, that must usesmaller time than users retriev models in full model databases. Clusteringanalysis technology often deals with the higher dimensional and greatcapacity for liquor data.For example,there are some media data ofimage,sound,3D model.But the increace of the dimension leads to the worseclusering result.So cluster analysis techniques should work together withother techniques.For instance,the dimensionality reduction technique canreduces the side-effect of the dimensionality curse;the sampling techniqueshavebeenappliedtoincreasetheefficiencyoftheclusteringalgorithm.We have design the model databases with the MySQL that is the opensource software.We solve some problems that are access the databases,changingthecharactersandstoragethebigobject.At the end, we test the retrieval system with the model data fromPrinceton benchmark databases.We have obtain the accuracy and effect withWeka that is another open source software.Weka is the best software in dataminging field.And we draw the curve of Precision-Recall.We can makeconclusion that the cluster technique can heighten t the retrieval efficiencyandrelevancefeedbacktechnologycanheightertheretrievaleffect.