A Study On Lmage Retrieval Based On Fractal Method

The fractal coding uses the compact Iterative Function System (IFS) to denote the image, so it has a good effect on the image compression, and it can be used in content-based image retrieval too. Huge video information systems need feasible retrieval methods urgently now. To improve the low efficiency in other methods, this thesis develops several image retrieval methods by different retrieval characteristic based on the fractal coding.The first method: The retrieval method based on the IFS There are similarities in different images, while the degree of the similarity is different. To get the IFS of the coding image is to make the sub-block in the retrieval image self-affine-contraction to approximate the sub-block in the images of the image database in this thesis. The PSNR(Peak Signal and Noise Ratio) which is computed by the decoding image and the coding image is used as the retrieval measurement. The experiments have proved this method can retrieve the different-pose-and-expression images of the same person. But the change of pose and expression can not be too much, otherwise the retrieval efficiency is low.The second method: The method based on a pair of parameters (D,LMSE)of a range blockThe fractal code of a range block is one to one correspondent to a domain block and the LMSE . By using the pair of parameters correspondent to a range block as the retrieval characteristic, the retrieval velocity is highly increased, and the retrieval time isapproximately 1/8 of that in the first method. But different-pose-and-expression images of the same person can not be retrieved by this method.The third method: The method based on a pair of parameters (D,LMSE)of all range blocksIn fractal coding, the range block is equal to a pair of parameters which are correspondent with the domain block and the Least Mean SquareError( LMSE ), as R5, (Df,LMSE,) .The pairs of parameters whichcorrespondent to all range blocks are used as the retrieval characteristic. The retrieval image is divided into range blocks the same as those stored in the database, then the range blocks are coded in terms of the preset domain blocks, which results in a LMSE. Compute the Euclid distance between the above LMSE and the one stored in the database. Average the Euclid distance of all the range blocks in the retrieval image, then the images whose average Euclid distance are less are the retrieval images. The experiments have proved that the images with different-pose-and-expression of the same person can be retrieved and the retrieval efficiency is the highest of all.