Research On Fast Fractal Compression Of Image Sequences

Image compression has always been a fundamental issue in the field of computer vision.Among them,fractal compression algorithm is a lossy image compression method.The algorithm has advantages of high compression ratio,reconstruction at any scale,and fast encoding.However,the existing fractal image compression algorithms are only for two-dimensional images,and there are few studies on three-dimensional images and two-dimensional image sequences.Due to the strong self-similarity of three-dimensional images and continuous image sequences,the use of fractal method to compress them has good theoretical and application prospects.Therefore,a method based on fractal for fast compression of continuous image sequences is proposed in this paper.First,the image sequence is segmented based on its spatial-temporal similarity.Image sequence is an unstructured data that has both time and space attributes.Before compressing the image sequence,it can be cut into independent fragments based on the spatial-temporal similarity.After color histograms of all the images are obtained,similarity of the histograms is calculated using the Barr distance.If the similarity between two frames of images is greater than a predetermined threshold,it is considered that a great change has occurred between the two frames of images,so the two are divided into different image fragments.Secondly,a self-characteristic of the image block is extracted.The calculation process based on the image block's own characteristic replaces the mutual calculation process for two types of image blocks in traditional fractal compression,thereby reducing its calculation complexity.Domain blocks are classified according to the characteristic of the image block,that is,domain block classes are divided by setting a threshold of the characteristic value.After obtaining all domain block classes,calculate the distance from each range block to the center of every domain block class.The class with the shortest distance is selected for matching search.Therefore,each range block only needs to find the best matching block in a certain class of domain block.Finally,this paper experiments the proposed method on two data sets respectively.(1)Take three standard grayscale image sequences of Walter Cronkite moving head,Chemical plant flyover(close view)and Chemical plant flyover(far view)as examples for testing.(2)Take an MRI image sequence in the ADNI dataset as an example for testing.Experimental results show that our method can not only effectively reduce the impact of compressed image sequences due to volume problems,but also solve the problem of too long compression time caused by too many domain blocks.