Based On The Support Vector Machine Image Retrieval Method Research

In recent years, with the popularity of digital electronic productsand the rapid development of information technology, digital imageprocessing has entered the high-speed development period. All walks oflife to the image of the more widely used, with the rapid growth of imageinformation, every day producing to a large number of digital images. Thedigitized information allows users to use more convenient. Since becomingmore and more a large image database, but if no valid image data management,a lot of information will be submerged in the database, when neededtraditional keyword-based image retrieval technology has been unable tomeet user needs, unable to retrieve out. In order to solve this problem,have been proposed content-based image retrieval technology itself. Withthe use of computer image classification techniques for quantitativeanalysis of images to replace the human visual interpretation, theestablishment of efficient, fast image retrieval. And on this basisdeveloping an image management system to help users store and retrieveimage database.The research project is based on SVM retrieval system, need to breakthrough the image feature extraction and classification of two keytechnologies.Studied in this paper is based on SVM classification methods. Basedon statistical learning theory SVM image classification method toclassify the image, can be achieved automatic and independent studyclassification, is the current and future main directions of research inthis field. SVM method can successfully solve the small sample,high-dimensional and local minimum problems. The goal is to get the optimal solution under the existing information, not just the number ofsamples tends to infinity the optimal value.Image retrieval include color, shape and texture in the visualcharacterization and extraction algorithm, similaritymeasure,normalization etc. And based on the histogram further researchof Improved grayscale histogram area map Retrieval method. Then bystudying the SVM theory and technology, construct multi-class classifierof the SVM, implemented a SVM-based image classification and retrievaltest system, and based on SVM for relevance feedback, the feedback problemof insufficient of samples for the algorithm to improve and achieve.Experiments show: based on SVM pre-classified in this image database onthe basis,an integrated feature extraction algorithms and relevancefeedback algorithm to retrieve the results obtained to better meet userrequirements in this paper.Technology and ideas of the system frameworkcan be extended to other media repository applications, with research andapplication of double significance.SVM-based image retrieval method of the basic process, includingimage preprocessing, image feature extraction, image classification, theoutput results and so on.