Research On Post-processing Techniques In DCT Compressed Image Communication

Sound, text and image are three kinds of information, which are oftenencountered in daily life. Because the image information has stronger characteristic ofintuition than that of others, it has been the main information source for people toacquire information nowadays. Along with the fast development of moderncommunication and computer techniques, the requirements of image information areincreasing. So the image communication becomes one of the most important ways totransmit information.However, for the hugeness of the image information, it must be compressedbefore being transmitted in real-time through the band limited channel. Theblock-based discrete cosine transform (DCT) is a fundamental component of manyinternational image and video compression standards. The distortions appearing in thedecoded image are inevitable, during the high compressed image informationtransmitted through the error-prone channel. Quantization distortion and transmissiondistortion are two main causes for compressed image communication based on DCT.In this dissertation, two kinds of post-processing techniques, namely blockingartifacts reduction and error concealment, are investigated and performed to reducethese two kinds of distortions to improve the quality of decoded image. The maincontributions of this dissertation are concluded as follows.Firstly, Markov Random Field (MRF) model for digital image and imagerecovery method based on Maximum a Posterior (MAP) are presented in the preface.Under the MRF-MAP framework, the Iterated Conditional Modes (ICM) was used tosolve the cost function. The model of anisotropic diffusion and projection ontoconvex sets (POCS) for image enhancement and recovery are also introduced in thepreface.Secondly, The reason causing blocking artifacts in BDCT compressed image isanalyzed, and then four kinds of post-processed de-blocking algorithms are presented.The first one is an adaptive de-blocking algorithm based on MRF. A visibilityfunction of blocking artifacts is introduced, then a new adaptive potential function isproposed using the visibility of the blocking artifacts and edges. The secondalgorithm takes advantage of anisotropic diffusion equation, which is constructedbased on the image local content. To avoid excessively smoothing texture and edgeinformation, a diffusion speed parameter is employed. Diffusion slows down whileprocessing texture regions, and speeds up while processing smooth regions. The thirdalgorithm is based on human vision system (HVS). For smooth regions,one-dimension DCT domains filter is applied in reducing the blocking artifacts. Forthe texture regions, a spatial filter is applied in doing so. In the fourth algorithm,blocking artifacts are modeled as step functions and then the image blocks are dividedinto three categories, i.e., smooth blocks, texture blocks and edge blocks. For smoothblocks, the expression of amplitude of blocking artifacts is deduced firstly. And thenan adaptive smooth filter is introduced according to the amplitude of blockingartifacts and smooth degree function. For texture blocks and edge blocks, the Sigmafilter is used to smooth the block boundaries. Among the above four de-blockingalgorithms, the first two ones belong to spatial iterative algorithms, while the last twoones belong to transform domain combined spatial domain non-iterative algorithms.Thirdly, the transmission error often happens in image transmission through theerror-prone channel. The loss of a single bit often results in the loss of the wholeblock or several consecutive blocks, which seriously affects the visual quality ofdecoded images at the receiver. In the circumstances of intra-coded mode, spatialerror concealment algorithm is presented to improve the quality of corrupted image.The missing blocks are classified into uniform blocks and edge blocks with the edgeinformation extracted from the surrounding correctly received blocks. For the missinguniform blocks, they can be concealed by the simple linear interpolation. For themissing edge blocks, the initial values of the missing blocks can be obtained byprediction based on gradient adaptive prediction (GAP). MRF-MAP model is used toimplement the optimization with the initial values of the missing blocks. In thecircumstances of inter-coded mode, two error concealment algorithms are presented.One is motion vector recovery algorithm based on MRF. The motion vectors field ismodeled as Gauss-Markov Random Field, and the weight is selected adaptively basedon the spatial information and temporal information. The other is hybrid errorconcealment. The missing image blocks are divided into lower active ones and higheractive ones. For the lower active blocks, average motion vector algorithm is used torecover the missing image blocks. For the higher active blocks, POCS method is usedto recover the missing image blocks.Lots of simulations show that the proposed post-processing algorithms caneffectively remove the quantization distortion and transmission distortion for BDCTcompressed image communications.