【作者】 石敏；

【导师】 谢胜利；

【作者基本信息】 华南理工大学 ， 通信与信息系统， 2005， 博士

【摘要】 随着计算机技术和通信技术的迅速发展,图像压缩编码问题一直是研究的热点。不断地探索和研究高压缩比,高质量,实现简单的图像压缩算法仍然是一个重要的课题。基于分块DCT变换的压缩编码技术是已有图像和视频压缩标准,但在低比特率条件下,解码图像不可避免出现了方块效应,研究有效的去除块效应算法也是十分必要的。近年来,由于小波变换在时域和频域同时具有良好的局部特性,因而成为现代图像研究领域的热点之一。本论文将围绕小波变换技术在图像压缩编码方面的应用展开研究,同时对低比特率下的块效应消除算法进行了深入的探讨,具体内容如下:1、提出一种新的降低比特率的预测矢量量化图像编码方法。在编解码端设置大小不同的两种码书,首先基于方差把所定义的块进行分类,对于平滑块使用较小的码书去量化输入矢量。对于边缘块,就用较大的码书去量化输入矢量。还提出了一种有效的码书设计方法以提高生成码书质量。新的矢量量化方法运算速度快,在同样图像质量下降低了传输过程中所需的比特率。2、提出了一种基于小波变换的自适应快速码字搜索算法。该算法首先为输入矢量选择合理的初始匹配码字,然后利用多控制点的三角不等式和小波域中矢量的性质去排除不可能匹配的码字,最后通过搜索空间的逐次降低找到与输入矢量最匹配的码矢量。在需很少的预先计算量和额外存储量条件下,所提出的算法编码质量和全搜索相同,但是计算量却有极大的降低。3、将DPCM变换与整数小波变换相结合来实现图像的无失真压缩。在压缩算法中,首先对图像进行DPCM预测,将差值图像经过整数小波变换,然后再用层次数中集划分(SPIHT)算法进行无损量化,最后再经过相应的逆变换即可以得到重构的无失真图像。该方法简单有效,硬件实现方便。4、提出了一种基于DCT域的高压缩图像去块效应算法。该算法充分利用了人类视觉的掩盖特性,建立了块效应模型并给出一个简便的检测边缘标准,对于平滑块,对影响块效应的系数进行修正并用线性函数块代替阶跃函数块去消除块效应,最后再对处理后的平滑块和纹理块在DCT域中进行后滤波。该算法在不同的比特率下,对不同的图像,都能得到很好的去除块效应的效果。5、基于小波域特征分析提出了一种去除块效应算法,通常图像空域中的块效应,在小波域中的高频子带表现为平行线效应,竖直线效应或者网格效应。通过自适应操作算子去抑制这些效应,使得空域中块效应得到消除。新的算法运算速度快,能够很好的消除图像的块效应并保护图像的边缘。更多还原

【Abstract】 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.更多还原