Image Post-processing Techniques In Image Communication

Recently, image communication or transmission becomes one of themost important ways to transmit the information. But, due to the poorperformance of the nowaday channels for image transmission, thedistortions appearing in the received image are not avoided. The two maincauses for image distortions are transmission error and quantizationerror. Aiming to improve the image quality, the damaged image ispost-processed in the decoder side in this dissertation. For two kindsof distortions, error concealment and blocking artifacts reduction areeffectively performed, under the condition that the recovered image mustpreserve the original contents and satisfy the human visualcharacteristics.Firstly, under the framework of maximum a posteriori estimation, theimage post-processing is mathematically represented in the text. We sumup the deterministic and stochastic models for the priori distributionof the original image, and introduce the robust statistics for dealingwith the outliers in the energy function optimization process.Secondly, for the circumstance of losing image information causedby transmission error, four error concealment algorithms are proposed.The first one is a new algorithm based on Markov Random Field -Maximuma Posteriori. We use Discrimination Analysis to construct pixel-membership, and introduce the adaptive edge threshold and slope for Huberfunction to improve the performance. The second one is based on robustoptical flow. Because of the good performance at estimating the object'smotion, optical flow technique is used to recover the lost blocks in apixel-wise manner. The third one, median of selective motion vectorsalgorithm (MSMV), select and preserve all motion vectors of neighbouringblocks, which belong to texture blocks, and use the median value as themotion vector estimation of the lost block. The fourth one, weightedboundary matching algorithm (WBMA), considers the influence of edges onboundary matching. It uses the edge magnitude of all pixels on theboundary of the lost block and its neighbouring blocks to design weightedvalues, and then implements the weighted matching.Thirdly, for the circumstance of blocking artifacts caused byquantization error, four blocking artifacts reduction algorithms arealso proposed. The first one is based on MRF-MAP. It uses the maskingeffect for blocking artifacts caused by human vision system, andintroduces a blocking artifacts just noticeable function to adaptivelyadjust Huber function. The second one is based on improved line process,which seperates the blocking artifacts from real edges. It designs ablocking artifacts measurement (LPBM), and reduces the blockingartifacts by optimizing the object function. The main idea of the lasttwo algorithms is to process the blocking artifacts in smooth and edgeblocks (or homogenerous and inhomogenerous blocks) discriminatingly.The third one is an algorithm based on fast detection (BARFD), whichemphasizes a convenient and fast way to detect the edge blocks. The fourthone is an adaptive algorithm (ABAR), which emphasizes using an adaptiveSigmoid function to reduce the blocking artifacts for each row of thehomogenerous block.Simulation results show that the proposed error concealment andblocking artifacts reduction algorithms have good performance onrecovering the distortions and preserving the original characteristics(for exmaple, edges).