Tsai, M. J., Villasenor, J.D., Chen, F., "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, Vol. 6, No. 5, October 1996, pp. 519-521  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... or who has attempted to build an arithmetic encoder in hardware) Another of the techniques listed, called the EBCOT algorithm, has been chosen as the basis of the JPEG 2000 standard and is discussed further in [2] Not included as a successor to EZW is a technique called stack run coding (SRC) [44], which has low complexity and good coding performance. SRC is somewhat of a hybrid between early and modern wavelet coders since it resembles EZW in some aspects (same uniform quantizer for all subbands) and subband coding in others (no interband prediction) A comparison of the performance of ....

M. Tsai, J. Villasenor, and F. Chen, "Stack-run image coding," IEEE Trans. Circuits Syst. Video Technol., vol. 6, pp. 519-521, Oct. 1996.

....and more of the embedded bitstream. 2. 2 Set Partitioning in Hierarchical Trees (SPIHT) SPIHT is a fully embedded image compression technique with precise rate control and low complexity [6] SPIHT outperforms other image compression techniques such as JPEG [7] vector quantization, and stack run [8]. SPIHT is a refinement of the Embedded Zerotree Wavelet al..gorithm [5] It orders data progressively; i.e. globally important data is sent first. The decoder can stop decoding at any point in the decoding process and a lower quality image can be decompressed and reconstructed. SPIHT employs a ....

M.-J. Tsai, J. D. Villasenor, and F. Chen, "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, vol. 6, no. 5, pp. 519--521, October 1996.

.... remote surveillance, we consider a variable bit rate compression schema based on a multiple pass multiplealgorithm strategy in order to take advantages from the application constrains (a certain tolerance to the information loss) using both lossy and lossless compression mechanisms [5] 6] [7], 8] More formally called #m the compression rate after m interaction of the data compression algorithm, we assume that: #m = g 1 g 2 m = 1 m m 1 m 1 (4) where g 1 and g 2 are constant values. The figure 4 shows a possible curve for #m with g 1 = 0.1 and g 2 = 0.8. The total ....

Tsai, M. J., Villasenor, J. D., and Chen, F. "Stack-Run Image Coding", IEEE Trans. CSVT, vol. 6, no. 5, Oct. 1996.

....coder [4] developed recently for the coding of ocean temperature datasets. This coder, termed successive approximation runlength (SARL) coding, combines successive approximation embedded coding in the form of bitplane coding with efficient runlength coding similar to that of [5]. In order to produce an approximately embedded bitstream for a dataset as a whole, we produce SARL codings of the data field and grid field independently, and then interleave the resulting bitstreams. The interleaving of the two bitstreams takes place in the creation of packets and assembling ....

M.-J. Tsai, J. D. Villasenor, and F. Chen, "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, vol. 6, no. 5, pp. 519--521, October 1996.

....different partitions) Nonetheless, making this choice results in a reduction in the algorithm s execution time by approximately a factor of eight in our implementation. In this paper, we consider two techniques as candidates for entropy coding: direct arithmetic coding [9] and stack run coding [10]. For long sequences, the former method typically gives results close to the zeroth order entropy, while the second is effectively a higher order entropy coder that exploits that fact that large runs of zeros are common in quantized high pass image subbands. Huffman coding is not a good candidate ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, vol. 6, pp. 519--521, 1996.

....coding [18] are Huffman coding and arithmetic coding, with the latter generally providing better performance. In this paper, we consider the use of two entropy coding methods: adaptive arithmetic coding, which is appropriate when the source statistics are initially unknown, and stack run coding [19], which combines arithmetic and runlength coding and exploits large runs of zero coefficients, as is common in many subimages. The design of the quantizer sets is an important parameter in determining the performance of the codec. All quantizer sets are comprised of 64 level symmetric uniform ....

.... and the sequence to be coded is relatively large, an arithmetic coder operating on individual coefficients can achieve a rate very close to the zeroth order entropy [18] A coder that exploits correlation between coefficients can do better than the zeroth order entropy and the stack run method [19] does this by combining run length coding with adaptive arithmetic coding on a small alphabet. To work well, however, this technique needs to see large runs of zero coefficients, implying large subbands. In determining which entropy coding technique to choose, we experimented with both arithmetic ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, vol. 6, pp. 519--521, 1996.

.... over code trees and bit assignments [427] Recently competitive schemes have demonstrated that separate scalar quantization of individual subbands coupled with a sophisticated but low complexity lossless coding algorithm called stack run coding can provide performance nearly as good as EZW [504]. The best wavelet codes tend to use very smart lossless codes, lossless codes which e#ectively code very large vectors. While wavelet advocates may credit the decomposition itself for the gains in compression, the theory suggests that rather it is the fact that vector entropy coding for very ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding," IEEE Trans. Circuits and Systems for Video Technology, Vol.6, pp. 519--21, Oct. 1996.

....filters. Searching for high performance wavelets for image coding has been an active research area. The 9 7tapped FB constructed in [2] has been the most widely used one in the image coding community. Recently, some even length FBs having good compression capability have been discovered [3] [4]. In [5] a general class of biorthogonal coiflet systems was constructed. The resulting FBs have odd lengths. Those of even orders were shown to be symmetric and have comparable compression performance to the 9 7tapped FB. However, those of odd orders are neither symmetric nor antisymmetric and ....

....short lengths, we found that the 22 14 tapped FB with the antisymmetric biorthogonal coiflet system of order (7,5) whose coefficients are listed in Table 1, achieves the best coding performance. Now we compare it with the 9 7 tapped FB in [2] the 6 10 tapped FB in [3] and the 10 18 tapped FB in [4], in terms of coding performance. The latter three FBs have been shown to be among the best FBs for image coding. In Table 2, we list the numbers of vanishing moments of the four wavelet systems, which are measures of their regularity [2] In fact, the latter three FBs were constructed via ....

M.-J. Tsai, J. D. Villasenor, and F. Chen, "Stackrun image coding," IEEE Trans. Circuits Syst. Video Tech., vol. 6, pp. 519--521, Oct. 1996.

....3.2 Compression Results The simulations were conducted by transforming and compressing 512 by 512 Lena and Barbara monochrome images. The SPIHT compression algorithm without the subsequent entropy coding was used (see [14] Our 3, 6, and 7 tap filters, the 9 7 (see [13] the 10 18 filters (see [12]) and two filters denoted A1 and A2 (see [23] were tested and compared in performance in SPIHT compression for these two images at several rates. The 9 7 and 10 18 are considered to be among the best practical FIR filters for compression and the A1 and A2 filters are considered to be the best ....

....8 tap linear phase filters presented below: 1 3 3 1] 0.0437 0.1000 0.4827 1.000 1.000 0.4827 0.1000 0.0437] In Table 1 we denote them by A1 and A2. Figures 7 and 8 contain the plots of the results. The simulation results show the steady superiority of our 7 tap filter and 10 18 filter from [12] over all the other filters. Next in line is the 9 7 filter of [13] followed by the 6 and 3 tap filters. The filters from [23] are at the end of the performance hierarchy and the smaller filter performs better than the longer one. We also show the pictures of the restored images compressed at ....

[Article contains additional citation context not shown here]

M. J. Tsai, J. D. Villasenor, F. Chen, "Stack-Run Image Coding", IEEE Trans. on Circuits and Systems for Video Technology, Vol. 6, pp. 519-521, Oct. 1996

.... over code trees and bit assignments [427] Recently competitive schemes have demonstrated that separate scalar quantization of individual subbands coupled with a sophisticated but low complexity lossless coding algorithm called stack run coding can provide performance nearly as good as EZW [504]. The best wavelet codes tend to use very smart lossless codes, lossless codes which effectively code very large vectors. While wavelet advocates may credit the decomposition itself for the gains in compression, the theory suggests that rather it is the fact that vector entropy coding for very ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding," IEEE Trans. Circuits and Systems for Video Technology, Vol.6, pp. 519--21, Oct. 1996.

.... over code trees and bit assignments [301] Recently competitive schemes have demonstrated that separate scalar quantization of individual subbands coupled with a sophisticated but low complexity lossless coding algorithm called stack run coding can provide performace nearly as good as EZW [357]. The best wavelet codes tend to use very smart lossless codes, lossless codes which effectively code very large vectors. While wavelet advocates may credit the decomposition itself for the gains in compression, the theory suggests that rather it is the fact that vector entropy coding for very ....

M.J. Tsai, J.D. Villasenor, and F. Chen, "Stack-run image coding," IEEE Transactions on Circuits and Systems for Video Technology, vol.6, no.5, pp. 519--21, Oct. 1996.

....quite a few of the coefficients are non zero. With fewer zerotrees, B s becomes quite high as for each non zero coefficient we have to code the exact position of its S bit. This results in a smaller remaining budget for B c , and hence a larger cutoff. We also tried the stack run coding approach [6], but found the new method (explained below) to be better (again, because of the specific block structure and the high bit rates) This new method also has the advantage that it avoids complex entropy coding techniques such as Huffman coding and arithmetic coding, and hence is extremely simple, ....

Tsai, M. J., Villasenor, J. D., and Chen, F. "Stack-Run Image Coding." IEEE Transactions on Circuits and Sysems in Video Technology, 6(5):519--521, October 1996.

....[9] 26.67 28.17 34.19 40.21 Infinite order [10] 21.06 22.38 28.01 33.95 Transform [11] 18.18 19.40 24.94 30.85 Biorth. 5 3 [7] dv = 1 18.52 20.02 26.04 32.06 Biorth. 9 7 [7] dv = 5 18.72 20.22 26.24 32.26 5 3 [12] dv = 1 18.49 19.99 26.01 32.03 9 7 [8] dv = 5 18.13 19.63 25.65 31.67 10 18 [13], dv = 6 18.46 19.96 25.98 32.01 Proposed 5 3, dv = 1 19.47 20.67 26.20 32.10 Proposed 9 7, dv = 5 20.18 21.42 26.96 32.80 Proposed 10 18, dv = 10 20.77 22.05 27.64 33.51 Table 1 shows the distortion rate performance of different systems when coding a Gaussian AR(1) source with correlation ....

....different systems when coding a Gaussian AR(1) source with correlation coefficient 0:95. Since it is assumed that pdf optimized scalar quantizers are used, the coding constants are given by c i = p 3 2 [4] for all i 2 f0; 1; M 1g. The performances of the three well known wavelets 10 18 [13], 9 7 [8] and 5 3 [12] are included in the table as well as the distortion rate function [9] optimal unconstrained length filter banks [10] and optimal transforms [11] Gain optimized, biorthogonal filter banks with linear phase [7] are also shown in the table. The gain for the first aliasing ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding," IEEE Trans. Circuits, Syst. for Video Technol., vol. 6, no. 5, pp. 519--521, Oct. 1996.

....of C 4 EXPERIMENTAL RESULTS Two Algorithms We present a brief description of two algorithmic realizations of our proposed methods. Both variants use a 4 level DWT with either biorthogonal 9 7 tap filter [1] for lowresolution images (such as QCIF video sources) or biorthogonal 10 18 tap filter [10] for highresolution still images. The way they differ is related to the method of zerotree map coding: PACC 1: hybrid ZT RLE of significance map PACC 2: conditional coding of zerotree map Except for the lowest frequency band which is coded separately using a DPCM like procedure, all bands are ....

M. J. Tsai, J. Villasenor and F. Chen, "Stack-run Image Coding", IEEE Trans. on CSVT, Vol. 6, pp. 519-- 521, Oct. 1996.

....that was not in the training set. In the table, the PSNR between the original and the blurred image is given, followed by the PSNR s between the original and the compressed restored images at 1.75 bits pixel using different wavelets. We haveused biorthogonal wavelet transforms from [15] 16] and [17]. The notation (r; s) in the table denotes a wavelet transform with r and s filter taps for the analysis and synthesis high pass filters, respectively. The PSNR improvements for the restored images range roughly from 0.2 to 1.75 dB. Figures 3 and 4 show the original and blurred images, ....

....of Figure 5 over Figure 4 is significant, especially considering that the restored image of Figure 5 is compressed at 1.75 bits pixel. The restored image also shows signs of modest super resolution, which is very encouraging. Table 1: PSNR results PSNR (dB) Blurred CDF [15] VBL [16] TVC [17] (2,2) 3,9) 3,9) 6,2) 11,13) 18,10) 22.55 24.28 22.74 23.00 24.00 23.51 23.27 6. ....

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding", IEEE Trans. Circ. and Sys. for Video Tech., vol. 6, no. 5, pp. 519--521, October 1996.

....also included Rate SPIHT arith. code SPIHT uncoded (bpp) enc dec enc dec 0.25 0.25 0.22 0.16 0.06 0.50 0.55 0.49 0.38 0.22 0.75 0.71 0.99 0.55 0.28 1.00 1.05 1.04 0.88 0.44 Table 2. CPU times for images compression with 120 MHz Pentium processor running under Windows 95. results of Tsai et al.[8] for stack run coding, because it has similar complexity to SPIHT and in this case uses the same Barbara image and wavelet transform. At the two lower rates of 0.25 and 0.50 bpp, the PSNR results are slightly worse than those of SPIHT, but they are not expected to compare as well at higher rates, ....

....because it was proposed as part of a video coding method. 3. AMPLITUDE AND GROUP PARTITIONING 3.1. Entropy coding methods Recently, there have been some notable advances in efficiency for low complexity coding of already quantized images. One such advance, mentioned earlier, is stack run coding[8], where run lengths of zeroes and quantized values are mapped into a very compact four letter alphabet code which is then further entropy coded with an arithmetic code. The efficiency depends on the presence of runs of zeroes, so requires a transform such as DCT or wavelet prior to quantization. ....

[Article contains additional citation context not shown here]

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding," IEEE Trans. on Circuits and Systems for Video Technology 6, pp. 519--521, October 1996.

....the suboptimal heuristic is thus the sum of the complexities of the two algorithms. 5 Experimental Results We apply our wavelet packet SFQ algorithm to a variety of input images used by the image coding community. Both the 7 9 biorthogonal filters of [3] and the 10 18 biorthogonal filters of [13] are used in our experiments. Adaptive arithmetic coding [14] is used to code the quantized wavelet packet coefficients. All reported bitrates, which include 85 bits that are needed to inform the decoder of the best wavelet packet tree (of maximum depth 4) are calculated from the real coded ....

M. Tsai, J. Villasenor, and F. Chen, "Stack-Run Image Coding", IEEE Trans. Circuits and Systems for Video Technology, vol. 6, pp. 519-521, October 1996.

....interpolating wavelet of order 4 [29, 30] which has the advantage of having filter taps that are dyadic rationals. Others 18 x=0 x Dead zone Figure 1.10 Dead zone quantizer, with larger encoder partition around x = 0 (dead zone) and uniform quantization elsewhere. are the 10 18 filters in [31], and the 28 28 filters designed with the software in [32] One promising new set of filters has been developed by Balasingham and Ramstad [33] Their design procedure combines classical filter design techniques with ideas from wavelet constructions and yields filters that perform better than the ....

M. Tsai, J. Villasenor, and F. Chen, "Stack-run image coding," IEEE Transactions on Circuits and Systems for Video Technology, vol. 6, pp. 519--521, Oct. 1996.

....is the Golomb Rice coder originally described in [6] and [7] that constitutes a key step in the LOCO I [8] lossless image coding algorithm. Golomb Rice codes are near optimal for exponentially distributed sources. Another motivating factor in the present paper is the stack run coder described in [11], which utilizes a 4 ary symbol alphabet to efficiently map runs and levels into a symbol stream that is further compressed using adaptive arithmetic coding. The stack run mapping can also be expressed using a structured tree, but one which is quite different from the tree representing a ....

....2 respectively. Note that at each level in the Golomb Rice tree there are 2 k code words. The pdfs to which these trees are matched are shown in Figure 2, and as one would expect for a coder matched to exponential sources, fall off linearly when plotted on a semilog scale. In another publication [11], we have described a coding algorithm called stack run coding that performs well for quantized wavelet coefficients. Although the stack run coder was not discussed in [11] in terms of a prefix code, it utilizes a mapping which is equivalent to the tree shown in Figure 1 (c) in which the number ....

[Article contains additional citation context not shown here]

M.J. Tsai, J. Villasenor, and F. Chen, "Stack-run image coding," IEEE Transactions on Circuits and Systems for Video technology, vol. 6, pp. 519-521, October 1996.

No context found.

Tsai, M. J., Villasenor, J.D., Chen, F., "Stack-Run Image Coding," IEEE Transactions on Circuits and Systems for Video Technology, Vol. 6, No. 5, October 1996, pp. 519-521

No context found.

M.J. Tsai, J.D. Villasenor, and F. Chen "Stack-run image coding," IEEE Trans. on Circuits and Systems for Video Technology, vol. 6, pp. 519--521, Oct. 1996.

No context found.

M. J. Tsai, J. D. Villasenor, and F. Chen, "Stack-run image coding", IEEE Transactions on Circuits and Systems for Video Tech., vol. 6, pp. 519--521, Oct. 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC