基于模型的小波域的视频压缩编码
|
摘要
目前，许多实用的图像编码算法都是基于空间域的运动估计和补偿、预测误差的DCT及量化以及变换系数的熵编码的混合编码方法，如有关静止图像和视频图像的国际标准JPEG，MPEG-1，MPEG-2等。但基于DCT的算法有其固有的缺点，即方块效应，在压缩比较高时，图像质量会很差，因此人们一直在努力研究寻找更为有效的编码方法。小波变换由于它的时频局部性、适合描述非平稳信号以及它的人眼系统的特性和适应的特点，视频编码方面受到了越来越多的重视。
将所有的当前帧和参考帧均变换到小波域，然后在小波域直接进行运动检测和运动补偿。这种方法的运动估计时间相对于传统方法明显减少，解码图像也消除了块效应，得到峰值信噪比和主观质量都很好的解码图像。但是在块匹配算法中，全搜索计算量很大从而十分耗费时间。在多分辨率运动补偿时，对最小的子带的每一个块都做运动补偿，这样，由于小波空间方向树的结构，在低级子带的相应的位置上都要进行运动补偿，因此整个图像都要进行运动补偿，计算量很大，很浪费时间，并且搜索范围是固定的，这样在运动补偿时，在匹配精度和运动补偿时间之间就产生了矛盾。
本文通过运动检测的方法提取得到的运动矢量和运动区域是基于最小的子带的，利用三帧运动检测算法提取到图像的运动区域，然后将运动区域分成2×2大小的小块，根据P帧和B帧的不同，使每个小块在相应的图像上进行搜索，得到的运动矢量来代替多分辨率运动补偿时固定范围内搜索得到的运动矢量，这个时候的运动矢量是只针对运动区域的，由于头肩序列相临图像之间有很强的相关性，运动部分不多并且运动范围很小，这样只是很少的块进行运动搜索，而且在多分辨率运动补偿时，高分辨率的运动矢量由最低频子图像搜索得到的运动矢量来预测，所以这个运动矢量的精确与否是很重要的。因此则在搜索的时候搜索范围设得相对大些，这样得到的运动矢量就很准确，实验结果表明，根据本文方法实现的视频编码系统确实达到了满意的效果。
Recently, many applied image coding algorithms are mixed coding methods based on the spatial motion estimation and compensation, the DCT and quantification of displaced frame difference (DFD) and entropy coding such as some international standards about still images and video compression coding:JPEG, MPEG-1, MPEG-2 and so on. But the algorithms based on DCT have their inherent disadvantage that is "blocking effect" . The quality of the restored images is too bad at the high compression ratio. So much more effective coding methods have been gone for. Having the good characteristics of temporal-frequency localization and so on, wavelet transform is very suitable for describing nonstationary signal and adapt to man' s vision system. Thereby wavelet transform is increasingly concerned in the field of video compression coding.
After wavelet transform ,The current frame and the reference frame are performed motion detection and motion estimation, compared with the traditional method, this method uses little time about motion estimation, the block effect of image disappears after decoded , the decoded image has ideal PSNR(Peak Signal Noise Ratio and subjective quality , but in the block matching algorithm , the full search uses too much time . In the MRME(Multi-resolution Motion Estimation) , the smallest subband is performed motion estimation, thus, because of the structure of wavelet, the corresponding position is performed motion estimation, the whole image must be performed motion estimation , thus the computing time is very large, moreover the range of search is constant that makes motion estimation contradict with matching precision.
Motion vector and motion region gained after motion detection is based on the smallest subband, the Motion region is required by moving object detection using three difference method, it is divided some of blocks , according to the I frame and P frame, the block performs motion estimation in corresponding image, obtained motion vector replace that is obtained by constant range in MRME, thus the motion vector is based on motion regions, because of the high correlation of head-shoulder image, the motion regions are few and the range of motion are small, thus only a few of blocks are performed motion search. In MRME, the motion vectors of low frequency are predicted by that of high frequency, so the motion vectors
precision are very important, in searching the motion vector, the range of search range is comparatively large, so the motion vectors' precision are higher. The simulation results indicate that the performance of the video codec system using the algorithm proposed by author are satisfying.
将所有的当前帧和参考帧均变换到小波域，然后在小波域直接进行运动检测和运动补偿。这种方法的运动估计时间相对于传统方法明显减少，解码图像也消除了块效应，得到峰值信噪比和主观质量都很好的解码图像。但是在块匹配算法中，全搜索计算量很大从而十分耗费时间。在多分辨率运动补偿时，对最小的子带的每一个块都做运动补偿，这样，由于小波空间方向树的结构，在低级子带的相应的位置上都要进行运动补偿，因此整个图像都要进行运动补偿，计算量很大，很浪费时间，并且搜索范围是固定的，这样在运动补偿时，在匹配精度和运动补偿时间之间就产生了矛盾。
本文通过运动检测的方法提取得到的运动矢量和运动区域是基于最小的子带的，利用三帧运动检测算法提取到图像的运动区域，然后将运动区域分成2×2大小的小块，根据P帧和B帧的不同，使每个小块在相应的图像上进行搜索，得到的运动矢量来代替多分辨率运动补偿时固定范围内搜索得到的运动矢量，这个时候的运动矢量是只针对运动区域的，由于头肩序列相临图像之间有很强的相关性，运动部分不多并且运动范围很小，这样只是很少的块进行运动搜索，而且在多分辨率运动补偿时，高分辨率的运动矢量由最低频子图像搜索得到的运动矢量来预测，所以这个运动矢量的精确与否是很重要的。因此则在搜索的时候搜索范围设得相对大些，这样得到的运动矢量就很准确，实验结果表明，根据本文方法实现的视频编码系统确实达到了满意的效果。
Recently, many applied image coding algorithms are mixed coding methods based on the spatial motion estimation and compensation, the DCT and quantification of displaced frame difference (DFD) and entropy coding such as some international standards about still images and video compression coding:JPEG, MPEG-1, MPEG-2 and so on. But the algorithms based on DCT have their inherent disadvantage that is "blocking effect" . The quality of the restored images is too bad at the high compression ratio. So much more effective coding methods have been gone for. Having the good characteristics of temporal-frequency localization and so on, wavelet transform is very suitable for describing nonstationary signal and adapt to man' s vision system. Thereby wavelet transform is increasingly concerned in the field of video compression coding.
After wavelet transform ,The current frame and the reference frame are performed motion detection and motion estimation, compared with the traditional method, this method uses little time about motion estimation, the block effect of image disappears after decoded , the decoded image has ideal PSNR(Peak Signal Noise Ratio and subjective quality , but in the block matching algorithm , the full search uses too much time . In the MRME(Multi-resolution Motion Estimation) , the smallest subband is performed motion estimation, thus, because of the structure of wavelet, the corresponding position is performed motion estimation, the whole image must be performed motion estimation , thus the computing time is very large, moreover the range of search is constant that makes motion estimation contradict with matching precision.
Motion vector and motion region gained after motion detection is based on the smallest subband, the Motion region is required by moving object detection using three difference method, it is divided some of blocks , according to the I frame and P frame, the block performs motion estimation in corresponding image, obtained motion vector replace that is obtained by constant range in MRME, thus the motion vector is based on motion regions, because of the high correlation of head-shoulder image, the motion regions are few and the range of motion are small, thus only a few of blocks are performed motion search. In MRME, the motion vectors of low frequency are predicted by that of high frequency, so the motion vectors
precision are very important, in searching the motion vector, the range of search range is comparatively large, so the motion vectors' precision are higher. The simulation results indicate that the performance of the video codec system using the algorithm proposed by author are satisfying.
引文
[1] 侯自强．数字视频技术进展．中国图像图形学报，1996，1(1)：58-63．
[2] 何梦林，王莉艾．图像通信中的信息压缩技术[J]．数字通信，1996，16(1)，27-29．
[3] 李春华，张雨生，戚银城等．小波变换在图像压缩研究中的现状和趋势[J]．华北电力大学学报．2001．4，28(2)：87-91．
[4] Castleman K R. Digital Image Processing. Prentice Hall, 1996.
[5] Jacquin A E. Fractal Image Coding: A Review. Proc. IEEE, 1993, 81(10):1451-1465.
[6] 李杰，付萍．分形及分形图像编码．长春大学学报．1999，9(4)：5-8．
[7] Nasrabadi NM, King RA. Image coding using vector quantization: a review[J].IEEE Trans. On Communacation,1988,36(8): 957-971.
[8] Guan L, Kame IM. A vectordistribution model and an effective nearest neighbor or search method for image vector quantization[J].IEEE Trans. On Image Processing, 1997,6(12): 1688-1691.
[9] Rinaldo r, Calvagno G. Hybrid Vector quantization for multiresolution image coding[J].IEEE Trans. On Image Processing, 1997,6(5):753-758.
[10] Rabbani M. Jones P. Digetal Image Compression Techniques[J].SPIE, 1991,1450: 116-128.
[11] Witten IH, Neal R, Cleary J.Arithmetic Coding for Data Compression[J]. Communications of the ACM, 1987,30(6):520-540.
[14] Jacquin A E. Image coding based on a fractal theory of iterated contractive image transform actions[J].IEEE Trans. On Image Processiong, 1992,1(1);18-30.
[15] Wohlberg B, Jager G de. Fractaldcoding performance for first-order Gauss-Markov models[J].Electronics Letters, 1996,32:441-442.
[16] 沈兰荪．图像编码与异步传输．人民邮电出版社，1998年．
[17] Hotter M, Thoma R. Image Segmentation Based on Object-oriented Mapping Parameter Estimation. Signalprocessing, 1988,15(3):315-334.
[18] Mallat S G. Multifrequency Channel Decomposition of Images and WaveletModels [J].IEEE Trans. On Acoustics, Speech and Signal Processing, 1989, 37 (12): 2091-2110.
[19] Mallat S G. A Theory for Multiresolution Signal Decomposition: the Wavelet Representation[J]. IEEE Trans. On Pattern Recognition and Machine Intelligence, 1989,11(7): 674-693.
[20] Shapiro J M. Embedded image coding using zerotree of wavelet coefficients. IEEE Trans. Signal Processing, 1993,41(12):3445-3462.
[21] 刘贵中、邸双亮．小波分析及其应用．西安：西安电子科技大学出版社，1992．
[22] Wallace G K. The JPEG still picture compression standard. Comm ACM, 1991,34(4): 30-34.
[23] ITU-T. Video coded for audiovisual services.Recommendnadation H. 261, SGXV Report 95,1992.
[24] 李德健等．JPEG2000标准的特点及应用．电视技术，2001(1)：22-23．
[25] A. Said and W.A. Pearlman. A new、fast、and efficient image codec based on set partitioning in hierarchical trees, IEEE Trans. Circuits Syst. VideoTechnol, vol. 6, June 1996. pp. 243-250.
[26] A. Zandi, J. Allen, E. Schwartz, and M. Boliek, CREW:Compression with reversible embedded wavelets, IEEE Data Compression Conference, Snowbird, UT, March 1995, pp. 212-221.
[27] 尚艳霞．基于小波域的视频压缩编码．大连理工大学硕士学位论文，2003．3．
[28] 王栓等．基于差分图像的多运动目标的检测与跟踪[J]．中国图像图形学报，1999，4(A)．
[29] Alan J Lipton, H Fujiyoshi, Raju S Patil. Moving target classification and tracking from real-time video [J]. IEEE Transactions on Workshop Application of Computer Vision,1998:8-14.
[30] Alireza Behrad, Ali Shahrokni, Seyed Ahmad Motamedi. A robust vision-based moving target detection and tracking system[EB/OL].http: //ligwww. epfl. ch/～ali.
[31] Jaewon Shin. Initialization of visual object tracker using frame absolute difference [EB/OL]. http: //ise. stanford, edu/class/ee392j/projects/shin-report. pdf.
[32] Jain R,Nagel H. On the analysis of accumulative difference of picture from image sequences of real world scenes[J]. IEEE Transactions PAMI, 1979:206—214.
[33] 陈朝阳，张桂林．基于图像对称差分运算的运动小目标检测方法[J]．华中理工大学学报，1998，26(9)．
[34] 李红艳．图像低信噪比小目标检测与跟踪算法研究[D]．西安西安电子科技大学，2000．
[35] C E Daniell, D J Kemsley, W P Lincoln, et al. Artificial neural networks for automatic target recognition[J]. Opt Eng,1992,31(12):2521—2530.
[36] Van der Auwera, G.Munteanu, A. Lafruit, G. Cornelis. Video coding based on motion estimation in the wavelet detail images. Acoustics, Speech, and Signal Processing, 1998. ICASSP '98. Proceedings of the 1998 IEEE International Conference on, May 1998 Vol. 5, pp:2801-2804.
[37] Podilchuk Christine I,Jayant Nikil S, Nariman Farvardin. Three Dimentional Subband Coding of Video. IEEE Trans. Image Processing, 1995, 4(2):125-139.
[38] R. Abut. Full search and tree searched vector quantization of speech waveforms Gray. Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '82, May. 1982, Vol. 7, pp:593-596.
[39] J. Jain and A. Jain. Displacement measurement and its application in interframe image coding. IEEE Trans. Commun.,1981,29(12):1799-1808.
[40] T. Koga, K. Linuma, A. Hirano, Y. Lijima, and T. Ishiguro. Motion compensated interframe coding for video conferencing, in Proc. NTCS1, Dec. 1981, pp:C9.6.1-9.6.5.
[41] Y. Zhang, S. Zafar. Motion-compensated wavelet transform coding for color video compression. IEEE Trans On Circuits and System for Video technology Sept. 1992, vol. 2, pp. 285-296.
[2] 何梦林，王莉艾．图像通信中的信息压缩技术[J]．数字通信，1996，16(1)，27-29．
[3] 李春华，张雨生，戚银城等．小波变换在图像压缩研究中的现状和趋势[J]．华北电力大学学报．2001．4，28(2)：87-91．
[4] Castleman K R. Digital Image Processing. Prentice Hall, 1996.
[5] Jacquin A E. Fractal Image Coding: A Review. Proc. IEEE, 1993, 81(10):1451-1465.
[6] 李杰，付萍．分形及分形图像编码．长春大学学报．1999，9(4)：5-8．
[7] Nasrabadi NM, King RA. Image coding using vector quantization: a review[J].IEEE Trans. On Communacation,1988,36(8): 957-971.
[8] Guan L, Kame IM. A vectordistribution model and an effective nearest neighbor or search method for image vector quantization[J].IEEE Trans. On Image Processing, 1997,6(12): 1688-1691.
[9] Rinaldo r, Calvagno G. Hybrid Vector quantization for multiresolution image coding[J].IEEE Trans. On Image Processing, 1997,6(5):753-758.
[10] Rabbani M. Jones P. Digetal Image Compression Techniques[J].SPIE, 1991,1450: 116-128.
[11] Witten IH, Neal R, Cleary J.Arithmetic Coding for Data Compression[J]. Communications of the ACM, 1987,30(6):520-540.
[14] Jacquin A E. Image coding based on a fractal theory of iterated contractive image transform actions[J].IEEE Trans. On Image Processiong, 1992,1(1);18-30.
[15] Wohlberg B, Jager G de. Fractaldcoding performance for first-order Gauss-Markov models[J].Electronics Letters, 1996,32:441-442.
[16] 沈兰荪．图像编码与异步传输．人民邮电出版社，1998年．
[17] Hotter M, Thoma R. Image Segmentation Based on Object-oriented Mapping Parameter Estimation. Signalprocessing, 1988,15(3):315-334.
[18] Mallat S G. Multifrequency Channel Decomposition of Images and WaveletModels [J].IEEE Trans. On Acoustics, Speech and Signal Processing, 1989, 37 (12): 2091-2110.
[19] Mallat S G. A Theory for Multiresolution Signal Decomposition: the Wavelet Representation[J]. IEEE Trans. On Pattern Recognition and Machine Intelligence, 1989,11(7): 674-693.
[20] Shapiro J M. Embedded image coding using zerotree of wavelet coefficients. IEEE Trans. Signal Processing, 1993,41(12):3445-3462.
[21] 刘贵中、邸双亮．小波分析及其应用．西安：西安电子科技大学出版社，1992．
[22] Wallace G K. The JPEG still picture compression standard. Comm ACM, 1991,34(4): 30-34.
[23] ITU-T. Video coded for audiovisual services.Recommendnadation H. 261, SGXV Report 95,1992.
[24] 李德健等．JPEG2000标准的特点及应用．电视技术，2001(1)：22-23．
[25] A. Said and W.A. Pearlman. A new、fast、and efficient image codec based on set partitioning in hierarchical trees, IEEE Trans. Circuits Syst. VideoTechnol, vol. 6, June 1996. pp. 243-250.
[26] A. Zandi, J. Allen, E. Schwartz, and M. Boliek, CREW:Compression with reversible embedded wavelets, IEEE Data Compression Conference, Snowbird, UT, March 1995, pp. 212-221.
[27] 尚艳霞．基于小波域的视频压缩编码．大连理工大学硕士学位论文，2003．3．
[28] 王栓等．基于差分图像的多运动目标的检测与跟踪[J]．中国图像图形学报，1999，4(A)．
[29] Alan J Lipton, H Fujiyoshi, Raju S Patil. Moving target classification and tracking from real-time video [J]. IEEE Transactions on Workshop Application of Computer Vision,1998:8-14.
[30] Alireza Behrad, Ali Shahrokni, Seyed Ahmad Motamedi. A robust vision-based moving target detection and tracking system[EB/OL].http: //ligwww. epfl. ch/～ali.
[31] Jaewon Shin. Initialization of visual object tracker using frame absolute difference [EB/OL]. http: //ise. stanford, edu/class/ee392j/projects/shin-report. pdf.
[32] Jain R,Nagel H. On the analysis of accumulative difference of picture from image sequences of real world scenes[J]. IEEE Transactions PAMI, 1979:206—214.
[33] 陈朝阳，张桂林．基于图像对称差分运算的运动小目标检测方法[J]．华中理工大学学报，1998，26(9)．
[34] 李红艳．图像低信噪比小目标检测与跟踪算法研究[D]．西安西安电子科技大学，2000．
[35] C E Daniell, D J Kemsley, W P Lincoln, et al. Artificial neural networks for automatic target recognition[J]. Opt Eng,1992,31(12):2521—2530.
[36] Van der Auwera, G.Munteanu, A. Lafruit, G. Cornelis. Video coding based on motion estimation in the wavelet detail images. Acoustics, Speech, and Signal Processing, 1998. ICASSP '98. Proceedings of the 1998 IEEE International Conference on, May 1998 Vol. 5, pp:2801-2804.
[37] Podilchuk Christine I,Jayant Nikil S, Nariman Farvardin. Three Dimentional Subband Coding of Video. IEEE Trans. Image Processing, 1995, 4(2):125-139.
[38] R. Abut. Full search and tree searched vector quantization of speech waveforms Gray. Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '82, May. 1982, Vol. 7, pp:593-596.
[39] J. Jain and A. Jain. Displacement measurement and its application in interframe image coding. IEEE Trans. Commun.,1981,29(12):1799-1808.
[40] T. Koga, K. Linuma, A. Hirano, Y. Lijima, and T. Ishiguro. Motion compensated interframe coding for video conferencing, in Proc. NTCS1, Dec. 1981, pp:C9.6.1-9.6.5.
[41] Y. Zhang, S. Zafar. Motion-compensated wavelet transform coding for color video compression. IEEE Trans On Circuits and System for Video technology Sept. 1992, vol. 2, pp. 285-296.