Mage Content Retrieval Based On The Determination Of Image Class Members

With the extensive use of computer graphics in daily life and production,the demand for image analysis,retrieval and classification management is increasing.Due to the non-structural features of the image information bring great difficultiesto the corresponding information processing.Therefore,the issuesrelated image content retrieval,has become one of the hot issues of common among the researchers in the computer information area.In specific image retrieval applications,it often need to analyses and extracts features in several images of the same kind of images to support the efficient and accurate image retrieval,which get representation of images characters.It is described that the theory and technology about conventional image characteristics in image often based on color feature,texture feature and shape feature,combined with color histogram,color correlation extraction method by circular area divided into block,color histogram analysis method,the color feature of the image were extracted;using gray level co-occurrence matrix to extractimage texture feature,do the quantification of image feature;using the methods bases on the contour and region to extractimage shape feature,and by these two methods,the image shape features were quantified calculation in this thesis.To solve the traditional image extraction method of color feature issues which are low efficiency and image noise exists,it is described that using quad-tree method improve the extraction method of the image's color featurein a fast way according to the different sizes of rectangular division in this thesis,thus improve the execution efficiency.Because of the lack of high efficiency and accuracy of traditional texture feature extraction methods,it proposes a texture feature extraction algorithm based on Contourlet transform in this thesis.Due to the image of the regular shape feature extraction method for identification of the irregular shape of the lack of accuracy,it proposes a method based on contour image shape feature extraction based on by wavelet descriptor algorithm is improved in order to realize the accuracy of image shapefeature extraction in this thesis.In addition,because of the method of using single feature image retrieval for different categories of imageslack of accuracy.Therefore,it continues to see the single feature image retrieval reached the accuracy as a standard weight design in this thesis.And then,compared to the single feature retrieval method,it designs a method based on multi-feature of weightedin the retrieval.Thus,the accuracy of image retrieval is improved.At the same time,based on the DS theory,the characteristic value of the low accuracy is eliminated,so that the retrieval accuracy has been further improved.Finally,in combination with the SVM classification principle,completed the linear classifier,kernel function and the design of relevant variables,the classification of image retrieval method based on SVM is established.It is based on the image content for target images to make similarity retrieval by using the unified standard which is "the retrieval accuracy" as the evaluation standard of all kinds of retrieval methods in this thesis.In the search results,the higher proportion of the expected image,then the retrieval accuracy is higher.Through the experiments,this kind of "SVM classification image retrieval with the improvement of feature","weighted multi-feature fusion retrieval" and "weighted multi-feature fusion based on the theory of the DS retrieval" are compared with the average retrieval accuracy.