Font Size: a A A

Study On The Key Techniques For The Control And Security Mechanism Of The Object-Oriented Multimedia Information System

Posted on:2006-01-16Degree:DoctorType:Dissertation
Country:ChinaCandidate:J XiaoFull Text:PDF
GTID:1118360155953691Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
With the development of Internet, more and more people can accessmass of multimedia data in a convenient and economical way. In recentyears, for the convenience of people's using multimedia data, study onmultimedia database by object-oriented is transforming to theapplication demand for it, involving not only the key technology suchas control and security, but also the study on optimizing the searchingmethods and searching algorithm. Gradual formulation and perfection ofthe multimedia content description interface——MPEG-7, has promoted thedevelopment of high-efficient content based multimedia search engine.Content based multimedia information retrieval refers to theautomatically analyzing of the multimedia and indexing the content bycomputers, instead of people's devotion to looking for the keywords.Content based retrieval of video data is a new research directiondeveloped following content based retrieval of image.Control and security mechanism are the important insurance for theapplication of object-oriented information system of the multimedia.This thesis has carried on research to relevant several items of keytechnology. The research results have been put into practice in scienceand technology development project of Jilin Province "Application ingovernment's government affairs information processing ofobject-oriented multimedia database". Performance of the system showsthat the control is reliable, secure, responding fast and searchingefficiently. The author's main research work is as follows:1. Distributed control technology is the comprehensive applicationof computer technology, communication technology, display technology andcontrol technology. Distributed control technology combined with theintelligent decision-making theories can efficiently realize managementand control of the object-oriented multimedia information system.Multimedia database of distributed intelligent control system isdistributed database structure. Different layer of the database systemhas different structure. All information is distributed in each differentworkstation. In the distributed intelligent control system, theinformation that will be used must be stored in the same controlworkstation. Adopt the database tool software to produce the databasefile. Then, change the database file into the form stipulated in SDCSdatabase according to various kinds of characteristics of information. The system also includes software for editing of flow menu and dynamicviewing display model of workflow. Flow menu formulation course can bedivided into three steps: the first one, drawing the suitable figure;the second one, connect the figure with the real-time database; the thirdone, produce the data file. Dynamic display module is the function moduleof SDCS system. According to the graphical data file (.GDB) and indexfile of dynamic figure (.RDB), it can display steadily and dynamicallythe workflow and control flow of SDCS system. Realize through revisingindex file of dynamic figure (.RDB). If other module hopes to change thedisplaying content, it must change the content of index file of dynamicfigure (.RDB). The purpose to do it in this way is to display the softwareand data will be baled and make it become a target, so in this way, displaymodule can make communication module, real-time database module,historical database module and other functions module up and form aintegrated system very easily. 2. One of the support subsystems of control and intelligentdecision-making is a content-based image feature extraction subsystem.Its realization algorithm includes such a series of abstractionalgorithms of characteristic image as content-based color feature,lamination feature, form feature, motion feature, etc. The applicationand study of image retrieval focus on the feature extraction. Colorinformation is regarded as the basic characteristic of the colored image,generally it is applied to the natural scene with comparatively abundantcolor while searching. Using the texture feature often has a betterperformance to retrieve the image with rich texture information. Insituation that the object outline is clear and the edge is obvious, itis the shape based feature that can result with good retrieval performance.To achieve an ideal performance, the system combines two or three kindsof features to finish the retrieving task.For the global motion, we adopt block matching algorithm at first tocalculate the observed value of the algorithm that is sparse motionvectors. Then estimate the model parameter of the global motion with theobserved value; for local motion feature, confirm some motion areaaccording to motion vector fields that global motion model and blockmatching algorithm try to get at first, and then use optical flow to drawdense motion vector fields to these areas. In the course of realization of block matching algorithm, we proposedfast searching adaptive algorithm, through analyzing the characteristicof the video data and various kinds of existing algorithm. It is basedon temporal and spatial correlation of motion of the video, predicatingthe reliability of the initial searching point, selecting the searchingmode and the adaptive stopping criterion. The experiment indicates thatthis algorithm can get nearly the same PSNR with the exhausted searchingalgorithm with speeding up the searching process. When estimate the global motion parameters, because of the affectof foreground object motion and random errors caused by the block matchingalgorithm in the low-texture area, there exists much novel data in thecalculated motion vectors. This thesis determines the credibility by theobserved value of the algorithm that is sparse motion vectors throughSAD and texture level, delete some novel data first, and then use theiterative least squares estimation method. Succeed in deleting data ofnon-global motion parameters from the estimated data set. Thus weobtained more accurate estimation of the overall motion model. While calculating optical flow fields with iterative method, becauseof the existence of the global motion and its large variation range, itwill need iterative process of a lot of times. This thesis resumes globalmotion model parameter by motion of the camera, thus compensates withglobal motion vectors first calculating and solving, not only hasaccelerated the iterative convergence rate but also reduced the errorof approaching the result. 3. Other one of the support subsystems of control and intelligentdecision is a subsystem that draws the feature of image based onclustering index. In this subsystem, the author has proposed a modifiedfuzzy C-means clustering index algorithm. It solves choice problem ofinitialized class effectively. At the same time, it has such functionsas the deletion, division with trends, amalgamating, merging, andmeddling in. The realization of the algorithm takes FastMap projectionclustering method. And it built a fast image retrieval system based onclustering index at the same time. Used the modified fuzzy C-meansclustering algorithm, the running time for indexing will not linearlyincreases with the increase of the scale and the dimension of the featureof the database. Compared with sequential scanning algorithm its speedhas been improved greatly. Comprehensive performance is superior to theexisting fuzzy C-means clustering algorithm. The high-dimensional eigenvectors affect the speed of clustering andsearching directly. Existed solution to the high-dimensional dataclustering makes use of the dimension reduction method. Project highdimensional data to two-dimensional or on three-dimensional space, thusachieve the visual goal. For the image denoted by high-dimensionaleigenvectors, dimension reduction will cause seriously information loss.According to the above consideration, in order to solve this problem,the system has proposed a kind of method based on dimension reduction,have not merely improved the speed of searching and cluster, theexperiment shows (recall ratio and precision ratio) the accuracy ofsearching the result is superior to classical Karhunen-Loeve transformdimension reduction method. It is an effective high-dimensional featuredimension reduction method. To single feature retrieval, because its restraint information isinsufficient, sometimes the retrieval result will not match with people'svision. It is an effective method of solving this kind of problem throughthe comprehensive features. But, to determine the weight betweendifferent characteristics or between different weights of the samecharacteristic is a very difficult problem. The system takes the feedbacklearning method while carry out the retrieval, tune the weights to achievethe best performance and make the retrieval match with people's visionfeature. At the same time, it can improve the performance of theretrieving flexibility and system performance. In this thesis weintroduce similarity measurement and many feature weight tuningalgorithm. 4. H. 3xx videoconference standard is combined with security demandof the information system of the multimedia, which designed a safemechanism and realized the system. The videoconference system is one of the real-time network multimediaapplications. Usually it refers to individual or group in two or moredifferent places, through transmission circuit and multimedia equipment,convey each other the voice, image and file materials, realize the instant...
Keywords/Search Tags:content-based image retrieval (CBIR), image feature extraction, clustering indexing, modified fuzzy C-means(MFCM), sequential scan algorithm(SSA), FastMap mapping algorithm(FMPA), system security.
PDF Full Text Request
Related items