Image Retrieval Based On Svm Feedback And Multi-feature Fusion

In recent years,content-based image retrieval has become a very active research topic in computer and related disciplines. Based on the analysis of the existing content-based retrieval techniques, this paper do the following research:Firstly, this paper study color features, discuss color space, color feature extraction methods and color matching algorithm, and compare retrieval results for different methods. The color space of RGB, HSV, HMMD are compared through experiments, and the features are retrieved through color structure descriptor . After that, it discusses the texture features of image, mainly introduce the texture feature extraction method of Gabor wavelet in the frequency domain and the co-occurrence matrix extraction method of texture features in the space domain, and compare the results of these two methods through experiments.The method based on a single characteristic can only retrieve part of the attributes of image, gives the limited description of the image contents, and is lack of adequate distinction information. This paper presents a new retrieval method which integrates color structure descriptor and homogeneous texture descriptor, and provides related discussions and experiments on this new method. Experiments proved that this new method could achieve a more satisfactory result while doing image retrieval.Following the discussion of the feedback technology and support vector machine technology, the paper presents a SVM (Support Vector Machine) based feedback technology, the multi- characteristics image retrieval technology with the integration of color structure descriptor and homogeneous texture descriptor. Substantive comparative tests have proved that this new technology can achieve better retrieval result than previous method.Lastly, the paper introduces the design and implementation of an Image Retrieval System prototype, followed by the conclusion of this paper and the forecast of the development of content-based image retrieval.