Research On Hybrid Wavelet Fractal Video Coding

The video is the most important medium of information. Because of its hugesize, video information is very difficult to be stored and transmitted. Along withthe continuous development of transmitting channels, store media and imagecoding etc., the technology of digital video has also been made a great progress.But it is still one of hotspots in this field to get more efficient video coding.Both wavelet and fractal are very important tools in video field for a long time.And the research on them has never stopped. Because of the significance of thewavelet transform and fractal theory, some work on the two tools about theoryanalysis and practice had been done in this paper. Then on this basis two hybridwavelet fractual vedio coding schemes were proposed and verified.Video coding is structured into two parts, intraframe coding and interframecoding.Motionless image coding algorithms are always chosen in intraframe coding. Ianalyzed several efficient algorithms of wavelet and fractal, and proposed a hybridwavelet fractal intraframe image coding algorithm.Three following algorithms in wavelet domain are analyzed and verified.⑴Embedded Zerotree Wavelet Encoding (EZW). The EZW algorithm canpractise embedded image coding well using the wavelet coefficients. It includesthe three procedures: zerotree prediction, encoding the important map withzerotree structure and gradually closing quantification. These EZW coders haveshown excellent distortion-rate performance with low computational complexity,while generating an embedded bit stream. This latter property enables one to sendimages in a progressive manner and to encode images at any target bit rate.⑵ Set Partitioning in Hierarchical Trees(SPIHT). The SPIHT algorithminherited the zerotree structure of wavelet coefficients in EZW, and the structure iscalled SOT (spatial orientation tree). Not only the zerotrees but also the rest treesare all treated as a set in this algorithm. Consequently it can get the highercompression rate than EZW.⑶Extended Set Partitioning in Hierarchical Trees (ESPIHT). The algorithm ofESPIHT includes three parts mainly: ①conducting the fast lifting wavelettransform to original intraframe images. ②giving the wavelet coefficients of thehigh frequency subband in wavelet image different sight weight value accordingthe human being's sight features.③encoding the intraframe image with SPIHTalgorithm using the extended zerotree structure. We can draw a conclusion fromthe trial that ESPIHT algorithm is more progressive in image coding quality andspeed than EZW and SPIHT.In the fractal domain, it analyzed the basic intraframe image fractal codingalgorithm. It based on IFS theorem and collage theorem, found one group of affinetransformation in the given image, which can compose an IFS whose attractorapproximate the given image, and then recoded the relevant coefficients. Indecoding stage, the reconstructed image is generated using the contractiontransformations and the relevant coefficients of IFS by iterations. In addition, italso analysed and verified an improved motionless image coding algorithm infractal domain in this paper, which has enhanced the performance of the intraframecoding to some extend through reducing the searching region of the domain block.Based on the research of the wavelet and fractal domain, an improved hybridwavelet transform fractal intraframe image coding algorithm was proposed in thispaper. It has a target to reduce the coding time and improve the reconstructiveimage quality. Thus in this project, the harr wavelet base was chosen and the affinesearching reign was shrinked. It has got an improved effect in the trial.The interframe coding of video images is divided into two parts. They are themotion estimation of the sequence image and encoding the difference of thesequence image methods. In recent years, many specialists in video field proposedso many improved methods of the motion estimation algorithm. But it did not getany much prositive progress yet in the performance of the whole videocompression coding. Thus in the video coding proposal of this paper, the motionestimation method selected classical ones and the basic diamond searchingalgorithm was chosen here. The image difference coding adopted wavelet zerotreecoding methods, and it had gotten more compression rate compared with thetraditional DPCM method.In chapter 5th, there are two video image coding schemes designed, namelythe video coding based on subband motion estimation and compensation, and thefractal predict coding based on the extended set partitioning in hierarchical trees.In the proposal of the video coding based on subband motion estimation andcompensation, the intraframe coding algorithm adopted the hybrid fractal waveletcoding algorithm in chapter 4th, and basic diamond searching algorithm was usedfor motion estimation. It used wavelet zerotree structure to encode the imagedifference. The whole remnant blocks were viewed as the image difference totransform when encoding the predicted frame image. For the interframe codingmacroblocks, their motion was estimated, their motion vector was achieved andrecorded. Then the remnant blocks were put into the corresponding positions ofthe difference image. The intraframe blocks coding used the motionless imagemethod, and the reconstruct blocks were achieved after decoding. The differenceblocks were gained by the original blocks minus reconstructive blocks, and itwould be the interframe coding blocks. By transforming and coding the differenceimage, then proceeding arithmetic coding, the encoding image would be gained.We can draw a conclusion from the trial that it achieved a rather excellent effect inquality and compression rate. But because of the computational complexity, thecoding time is rather long, and it still has a lot of shortage to overcome.In the proposal of the fractal predict coding based on the extended setpartitioning in hierarchical trees, each of the sequence frames was decomposedinto thirteen subbands which had different directions and resolutions in waveletdomain after the four level pyramidal wavelet transformations. There is fourlowest and three higher resolution subands in them. The intraframe coding adoptedthe method of extended set partitioning in hierarchical trees. The first frame ofvideo proceed the intraframe coding, and the rest frames adopted fractal predictcoding method and used the interframe wavelet zerotree to encode the motionvectors and compensation. After each frame finished the intraframe fractal predictcoding, it would then start decoding procedure with the wavelet coefficients of theformer frame in frame memory. The matching subtree did not need iterations withthe fractal decoding, so it could be finished in short time. Therefore, this algorithmsaved plenty encoding and decoding time. But after the trial, it was found that thereconstructive video image was barely satisfactory in subjective quality. It neededfurther research and betterment.