Application Of Fractal Coding For Image Retrieval

Fractal coding parameters can effectively represent essential features of images. Fractal coding,as a new image compression technique, has been applied into image retrieval. As we know, imageretrieval based on compressed domain is one of the hottest researches in content-based image retrieval.Since the decompression process is avoided, the computing complexity is low, that is, the real-time,efficiency and flexibility of image retrieval are improved. This technology is suitable for dynamicdatabase, Internet network retrieval and limited computing resources environment such as handheldnetwork terminal, which plays a very important role in network information security as well.The main researches of this paper are as follows:1. Do researches on orthogonalization fractal coding algorithm, and have verified that thedecoding speed is higher than that of the basic fractal coding. Meanwhile, collage error is proposed toretrieve images and the experimental results indicate that the collage error can represent essentialcharacteristics of images effectively.2. An effective image retrieval method based on fractal dimension using kernel densityestimation is proposed. Firstly, the fractal coding parameters are preprocessed, then the statisticalmethod—kernel density estimation is used to analyze the parameters. These statistical characteristicsare employed as retrieval indices. The experimental results show that this method has not only higherretrieval accuracy rate but also lower computing complexity than the existing methods such ashistogram method.3. Since bandwidth exhibits a strong influence on estimation results, a variable bandwidth isproposed according to the data distribution. Then the feature vectors of fractal parameters and collageerror are employed as retrieval indices. The experimental results indicate that retrieval accuracy rateincrease, computing complexity decrease as well.4. A robust index (rotation, translation, scaling invariance) extracted from fractal parameters isproposed, that is improved Hu invariant moment. The index is extracted from an approximate imageconstructed by mean range blocks. Then combines statistic characteristic of fractal parameters withthe Hu invariant moment index, the weighed indices are employed to compare the similarities amongimages. The experimental results show that the weighted indices perform better than a separate index.