Description Of Multiresolution Color And Texture Features Based On Complex Wavelet Theory For Cbir

As the mass storage facility of computer, multimedia and network technology rapidly developed and widely applied the image data increased greatly in the internet. In recent years, the technology that people how to retrieve images that they need from the image database quickly and exactly has become a focus. At the present time in the international, content-based image retrieval(CBIR) technology is one of the resolutions to this challenging question.First of all, the background and importance of the content-based image retrieval were illuminated in this part. Then, the situation and hotspot of the CBIR technology today were introduced. At the same time, the native application situation and overseas application situation of content-based image retrieval system were explained briefly. In the final of this part, some representative applications of content-based image retrieval system were listed.Secondly, the basic principles of content-based image retrieval system and some key technologies like similarity measure were presentation, and then the general framework was discussed in this section. The key factors that decide the efficiency of an image retrieval system are the description of image content and measure the similarity between two images.In the next part of this paper, the complex wavelet theory and its directional characteristics were introduced, here mainly focused on the dual tree complex wavelet transformation (DT-CWT), then compared DT-CWT with the traditional wavelet transformation in description of the directions information of image, the description of image characteristics in frequency domain were studied here also. Some methods discussed here on extraction of color and texture features of images. The retrieval method proposed in this paper will mainly use color auto-correlogram as the color features, and use the information of image blocks which can describe texture well as the texture feature, introduced two new operators to describe the texture characteristics, BDIP operator extracted the image edge and valleys information, BVLC operator describe texture roughness. Combination of these two features to represent the characteristics of image at last.Finally, based on the work of above, proposed a method of description and extraction image color and texture features in the dual tree complex wavelet transformation domain. This method convert the RGB color image into the HSV space first, then each component image of the HSV space is separately decomposed by DT-CWT, extracted the color features of component image from the decomposed H, S component, and texture features from the decomposed V component, at last, this two kind of features were combined and formed a joint image features vector to representation image characteristics for image retrieval. This section present the experimental of the whole retrieval technique proposed in this paper, and evaluate the algorithm with the precision versus recall method. Experimental results show that this method of extraction of features from DT-CWT domain can representation image better and had a good image retrieval result.