Research On Image Compression Encoding And Deblocking Algorithm At Low Bit Rates

With the fast development of computer and communication technology,image compression encoding is the hot research all along. It is importantto explore and study high compression ratio, high quality and easyimplementation for image compression algorithms. Block-based discretecosine transform (BDCT) technology is the international compressionstandard of image and video. The decoded image differs from visibleblocking artifacts at low bit rates. It is essential to study efficientdeblocking algorithms. Due to good local performance in both time domainand frequency domain, wavelet transform has become one of the most activeimage research fields. This paper mainly studies how to apply wavelettransform technology to image compression encoding application anddeeply discusses how to reduce blocking artifacts at low bit rates. Themain work is as follows:1. A new prediction-based vector quantization (VQ) method for imageencoding is proposed to reduce bit rates. Two codebooks with differentsize are employed at the encoder and decoder. Firstly, the defined blocksare classified based on variance. For smooth areas, the current vectorsare encoded with the small codebook. For edge areas, the current vectorsare encoded with the large one. An efficient method for codebook designis also presented to improve the quality of the resulted codebook. Theproposed method can really help to speed up the encoding time and reducethe bit-rate for the same image quality. 2. For the complexity of the encoding of vector quantization, anadaptive fast codeword search algorithm based on wavelet transform. Inthe algorithm, reasonablely initial codeword is chosen for the inputvector firstly. Triangle Inequalities using multiple control vectors andcharacteristics of vectors in transform domain are then utilized toreject non-matched codewords. The best-matched codeword is obtained byreducing search space gradually. The proposed algorithm significantlyreduces the computational complexity of VQ encoding without compromisingthe encoded image quality with little preprocessing and memory cost.3. A lossless image compression method combining DPCM transformwith integer wavelet transform is presented. Firstly, DPCM transform isused in the algorithm, and the difference image is transformed by integerwavelet. Then we can get the bit stream by lossless SPIHT algorithm.Finally the reconstruction image is obtained by corresponding inversetransform. The method is simple and hardware implementation is easy.4. A deblocking algorithm in DCT domain is proposed. Thecharacteristic of Human visual system (HVS) is sufficiently utilized inthe algorithm. The model of blocking artifacts is built and a convenientedge detection criterion is introduced. For smooth region, the parameterthat affects blocking artifacts is modified and the step block isreplaced with linear block. The DCT-domain post-filtering method isapplied to the updated block and texture region. The proposed method hasgood performance in reducing blocking effects at different bit rates fordifferent kinds of images.5. Based on the feature analysis in wavelet domain, a new deblockingalgorithm is proposed. The blocking artifacts in space domain refecthorizontal line-shape effect, vertical line-shape effect and grid-shapeeffect in high frequency subbands of wavelet domain. Adaptive operatorsare applied to suppress these effects to reduce blocking artifacts inspace domain. The new method is computationally efficient andsignificantly reduces blocking artifacts while preserving edge andtexture information.6. Based on wavelet domain and Markov random field(MRF), a newdeblocking algorithm is presented. The mathematical expression isdeduced to set the threshold for Huber function of MRF. The thresholdobtained by this expression can provide good tradeoff between protectingimage edge and improving image quality. The new algorithm can wellmaintain image edge information, and Peak Signal Noise Ratio (PSNR) isclose to projection onto convex sets (POCS) which has the best objectiveimage quality in total performance, while its speed is far more than thelatter. Meanwhile, the subjective visual effect of our alghorithm isclose to that of space methods.