Image Retrieval Based On Fractal Domains

Explosive increase of information content and networked informationcommunication are the prominent characteristic in the information age.Ifwe want to retrieval a mass of information ,we must have a high efficientand fast search engine.In this dissertation, we investigate thecompressed-domain image retrieval techniques,and select the challengingtask of image retrieval based on fractal domain to study. We mostly studythe algorithm of fractal codec according to the statistical property ofrange blocks and domain blocks,the theoretics basis of fractal index andimage retrieval entirly based on fractal domain.By summarizing the traditional fractal codec algorithm and analyzingthe statistical property of range blocks and domain blocks,we present theimproved fractal codec algorithm based on the statics property of blocksin this dissertation . The experimental result indicates that there arequicker coding speed and higher decoding image quality using the improvedalgorithm.Because not all of image features such as graylevel and histogram canbe extracted good indices,so we present the theoretics basis that fractalIFS(iterated function system) regards as the effective image feature toextract indices,i.e. fractal IFS has the property that the similar imageshave similar fractal IFSs,the dissimilar images have dissimilar fractalIFSs;the similar fractal IFSs have similar decoded images,the dissimilarIFSs have dissimilar decoded images.And we have proved the theoreticsbasis.In additon, because of the overhigh dimensions of fractalIFS ,fractal IFS can't be directly used as index to query image . Then,we present the method extracting the robust indices of the scale indexand Hu invariant moment index from the fractal IFS.To implement the FBIRS(fractal-based image retrieval system) inMicrosoft Visual C++ 6.0,we capture ten categories of images(about 2000images) to experiment. In the image retrieval ,image database is veryimportant and we select the technique of ADO(activeX data object ). Theexperimental result indicates that there is better effect using the prorate sum of two indices of scale index and Hu invariant moment index thanusing the single index of scale index or Hu invariant moment index ,andthese indices have been better tolerate-error of image rotation,imagetranslation and so on,which can effectively improve adaption of imagecompare search engine and speed of retrieval.