- PDF 免费下载

Size: px

Start display at page:

Download ""

纠糠拳向
7 years ago
Views:

1 DSP MATLAB TMS3267 SPIHT 67 DSP PI DT DSP 67 DSP +SPIHT + DSP I

3 Abstract Digital information, especially digital video and adio information, always has large amont of data which makes information storage and transportation difficlt. So it s necessary to research on and develop novel effective mltimedia data compressing method to solve this problem. Wavelet transformwt is a kind of domain transform method in image processing which takes hman visal and mental systems into consideration. By decomposing original image to mlti-resoltions in freqency domain, it provides ideal transform domain for image compression. The object of this article is to research on several wavelet transform based video compressing method and their rnning properties. We have done the following work: We have designed wavelet transform and zero-tree coding based still image compressing/decompressing programs in MATLAB environment. Zero-coding method is an effective scanning method of wavelet decomposed coefficients, which can eliminate correlation of wavelet coefficients in different resoltions. We have tested wavelet transform and SPIHT coding based still image compression method. SPIHTSet Partitioning in Hierarchical Trees method is a more effective wavelet coefficients scanning method originated from zero-tree method. We have designed programs in langage. We have tested mlti-resoltion motion estimationmrme techniqe to achieve mlti-frame compression in MATLAB. In MRME, coding of intra-frame and inter-frame are done in wavelet decomposition domain. MRME costs mch less time to perform motion estimation operation compared to block matching motion fll search algorithmfsa. We have applied wavelet transform and SPIHT coding based still image compressing programs on TMS3267 for hardware real-time compression. In order to enhance coding/decoding speed, we have rewritten image wavelet decomposition and composition program with 67 assembly langage in which we made se of parallel instrctions and pipeline optimizing skill to lessen calclating time. Image data are exchanged between host and DSP evalation board in a high speed throgh PI interface. Experimental reslts show when image compressing rate is high, decoded image of wavelet transform based compressing method can preserve objects otline well and has no block effect which may appear in DT block coding techniqe. Image compression rate can be changed to any given nmber by modifying parameter of the program. After properly designing, wavelet transform program on DSP has reached 67 s speed limit. As to MRME method, we only did simlation to get its resltant statistics becase of time limitation. Simlation reslts show reconstrcted image of this techniqe has good overall visal effect and has no block profile which may exist in block matching motion estimation method. MRME s time reqirement approximates that of block logarithm search algorithm. We conclded that designing DSP and II

4 wavelet transform + SPIHT coding + MRME method based video compressing system is applicable. If real-time video data collection part is added, we can get a perfect real time video collecting and compressing system. Keywords: Video data compression Wavelet analysis DSP techniqe III

5 Abstract DT JPEG 3.2 MPEG- 3.3 MPEG MPEG-4 4. Foier SPIHT MRME 6.2 MRME SPIHT DSP 7. SPIHT 7.2 SPIHT WT67PA DSP TMS3267 DSP I II IV

6 DSP S WT67PA SPIHT WT67PA WT67PA V

7 ITV VOD. B58mm 255mm 3dpi 2 /mm 6.6MB/ 65MB D-ROM 98 2 DA 44.kHz 6 / SIFSorce Inpt FormatNTS 4:4: IRInternational onsltative ommittee for RadioPAL 4:4:4

8 x y PM.2 VLSI IBM MPEG-2 IBM MOS-5L.5 MPEG- MPEG-2 IR6 5Mbps P P.3 DT 2

9 DT.4 DT JPEG MPEG- MPEG-2 MPEG-4 MPEG- SPIHT SPIHT SPIHT SPIHT matlab SPIHT DSP SPIHT DSP TI 67 SPIHT SPIHT DSP PI 3

10 2. DPM I DPM DPM DPM DPM DPM DPM DPM H.263 MPEG 4

11 5 5 H.263 MPEG / / 2.2 DT K-L K-L DT DT K-L DPM -6-4 DT 2 DT DT DT 8 5

12 6 2 DT DT 8 2 ] 6 2 cos 6 2 cos, [ 4, = = = π vπ y x y x f v v F x y 2 2 ] 6 2 cos 6 2 cos, ' ' [ 4, = = = π vπ y x v F v y x f x y = = = ',, 2 ', v v v

13 2.3 RL DT Hffman a a 2 a 3 a 4 a 5 a 6 a % 7

14 2.4 H.26 MPEG- MPEG-2 M BMA PRA 2-3 / 6-7 / 2-2 I Î e e I = I ˆ + e I I q = e e I Î e e I + - Î e Q P e + + I e P Q e h

15 k M NF MSE MAD MAD MAD MAD i, j f k K i j MAD MADi,j 2 MN M N = m= n= f m, n f k k m + i, n + j 2 3 9

17 Degree papers are in the Xiamen University Electronic Theses and Dissertations Database. Fll texts are available in the following ways:. If yor library is a ALIS member libraries, please log on and sbmit reqests online, or conslt the interlibrary loan department in yor library. 2. For sers of non-alis member libraries, please mail to [email protected] for delivery details.

2

10384 -------- 9601008 UDC---------------------- 1999 1999 1999 ------------------------ ------------------------ 1 2 .4...4 7 14.14..18..22..22..24 3 4.27.32.39 1 2 3 4 5 6 7 8 9 10 11 5 12 6 1 2 3 4