A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Proc. ICIP-94, vol. 3, pp. 856--860, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....are often used in practical coders (cf. 11] 22] The diversity and success of adaptive methods suggest an insensitivity of adaptive quantization. We observe in the backward and parametric adaptation, di erent models and estimation techniques give similar distortion performance (cf. 11] [13], 22] 24] and references therein) In [11] the quantizer estimate is updated on a pixel by pixel basis, an in nite mixture of quantizers in the words of the authors. In [22] pixel by pixel model parameter estimates are mapped, via a context variable, into classes of subband coecients de ning ....

....has been observed to have little impact on distortion performance ( 2] Nevertheless. we note that although a rough approximation to the source distribution may give adequate results for quantization purposes, better models are still important for ecient coding of quantized wavelet coecients. In [13] piece wise linear approximations of model densities are used, with competitive results. In [24] simulation studies show that even crude density approximations, far from accurately describing the true source distribution, induce close to optimal quantizer designs. Indeed, adaptive quantization ....

[Article contains additional citation context not shown here]

A. Ortega and M. Vetterli, \Adaptive quantization without side information", IEEE Trans. on Image Proc. vol 6, pp 665-676, 1997

....predictor (H i ) with the quantizer (c i ) i.e. they are jointly specified by one index. This coupling also allows for the predictor and quantizer to be jointly optimized in the codebook design phase. While techniques for adaptive coding without side information have appeared in the literature [36, 70, 85, 105], these methods base their adaptation on previously decoded signal; our approach utilizes the current signal. The decoder is able to follow this adaptation since it is explicitly described in the transmitted index. Vector (scalar) restoration requires the same bit rate as non adaptive PVQ (DPCM) ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Proc. IEEE Int. Conf. Image Processing, vol. 3, pp. 856-860, 1994.

....initial work on entropy coding (e.g. Huffman coding) relied on knowing, or measuring, the source distribution, adaptive schemes make no prior assumptions on the source statistics, which the coders try to learn. Here, we are concerned with designing 1 For publications related to this chapter see [75] 162 adaptive quantization algorithms which exhibit characteristics similar to those of the abovementioned adaptive lossless compression schemes. In the most general context we can define our problem as that of adapting some or all of the parameters of a quantizer entropy coder system (including ....

A. Ortega and M. Vetterli. Adaptive quantization without side information. In Int'l Conf. on Image Proc., ICIP'94, Austin, Texas, Nov. 1994. To appear.

....class of backward adaptive methods. In this mode, adaptation can be done at a finer (pixel) level since there is no side information, but the adaptation context must be causal. Examples include adaptive arithmetic coding [8] pel recursive coders [9] and recent backward adaptive lossy quantizers [10]. A fundamental question facing practical coders is: what is the best way to combine forward and backward adaptation A backward adaptive framework incurs no overhead cost but is vulnerable to unreliable causal contexts. A forward adaptive framework is more robust but the adaptation cost could be ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Int'l Conf. on Image Proc., ICIP'94, vol. 3, (Austin, Texas), pp. 856--860, Nov. 1994.

No context found.

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Proc. ICIP-94, vol. 3, pp. 856--860, 1994.

No context found.

Y. Yoo and A. Ortega, "Adaptive quantization without side information using scalar-vector quantization and trellis coded quantization," in Proc. 29th Asilomar Conf. Signals, Systems, and Computers, Pacific Grove, CA, Oct./Nov. 1995.

No context found.

A. Ortega and M. Vetterli, "Adaptive quantization without side information, " in Int. Conf. Image Processing, ICIP'94, Austin, TX, vol. 3, pp. 856--860.

....at the encoder as well as the cost of explicitly conveying the updated quantizer parameters via side information to the decoder. In backward adaptation systems like pel recursive coders [1] the decoder infers the adaptation parameters without the encoder needing to send any side information [2]. The price paid for this zero side information is, of course, the increased decoding complexity. In a number of image and video coding applications (e.g. broadcast, CD ROM storage or centralized processing applications like remote data base retrieval) the decoder complexity needs to be kept ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Int'l Conf. on Image Proc., ICIP'9J, vol. 3, (Austin, Texas), pp. 856-860, Nov. 1994.

....set of available quantizers the encoder can select the best quantizer for a given block or region of the image. Our goal is to demonstrate that backward adaptive quantization schemes are also useful in image compression. As a starting point we consider the adaptive quantizer introduced in [8, 9] which estimates the dis tribution of the data based on the previously quantized information. It has been shown that for image subbands it is useful to separate the coefficients in a given subband into different classes so that different quantizers can be applied to each class. We propose a novel ....

A. Ortega and M. Vcttcrli, "Adaptive quantization without side information," in Proc. ICIP'9J, vol. 3, (Austin, Texas), pp. 856 860, Nov. 1994.

....we will refer to as adaptive quantization. Early examples of adaptive quantizers [2, 3] focus on the dynamic range adaptation of a uniform quantizer based on past quantized data. Recent work has extended the adaptation to include updating of both dynamic range and bin sizes for a scalar quantizer [4]. This quantizer can learn the distribution of a stationary memoryless source and, therefore, performs well in the presence of long range dependencies of data as, for example, in a bimodal distribution. The goal of this paper is to use the adaptive quantization technique of [4] as a building ....

....scalar quantizer [4] This quantizer can learn the distribution of a stationary memoryless source and, therefore, performs well in the presence of long range dependencies of data as, for example, in a bimodal distribution. The goal of this paper is to use the adaptive quantization technique of [4] as a building block in quantization schemes, such as the scalar vector quan tizer (SVq) 5] and the trellis coded quantizer (TCq) 6] which are constructed based on an underlying scalar quantizer (USQ) Our motivation is to demon strate how adaptivity can be added and provide good results for ....

[Article contains additional citation context not shown here]

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Int'l Conf. on Image Proc., ICIP'9d, vol. 3, (Austin, Texas), pp. 856-860, Nov. 1994.

....at the encoder as well as the cost of explicitly conveying the updated quantizer parameters via side information to the decoder. In backward adaptation systems like pel recursive coders [1] the decoder infers the adaptation parameters without the encoder needing to send any side information [2]. The price paid for this zero side information is, of course, the increased decoding complexity. In a number of image and video coding applications (e.g. broadcast, CD ROM storage or centralized processing applications like remote data base retrieval) the decoder complexity needs to be kept ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Int'l Conf. on Image Proc., ICIP'94, vol. 3, (Austin, Texas), pp. 856--860, Nov. 1994.

....Laplacian parameter of a given class, based on past quantized data, thus allowing us to further take advantage of localized behavior. The on the fly update of the Laplacian parameter facilitates backward adaptation of the parametric quantizer in a way similar to the ideas introduced in [19] and [20] where non parametric piecewise linear distributions were used. Finally, to demonstrate the effectiveness of our approach we design a complete subband image coding system based on the uniform threshold quantizer (UTQ) This allows us to highlight the benefits of adaptation when comparing our ....

Y. Yoo and A. Ortega, "Adaptive quantization without side information using SVQ and TCQ," in the 29th Asilomar Conf. on Signals, Sys., and Computers, (Pacific Grove, CA), Oct.--Nov. 1995.

....Laplacian parameter of a given class, based on past quantized data, thus allowing us to further take advantage of localized behavior. The on the fly update of the Laplacian parameter facilitates backward adaptation of the parametric quantizer in a similar way to the ideas introduced in [19] and [20], where non parametric, piecewise linear distributions were used. Finally, we demonstrate how these two techniques can be included in a complete subband image coder using a uniform threshold quantizer (UTQ) 21] as the baseline quantizer. The resulting image coder shows very good performance, ....

Y. Yoo and A. Ortega, "Adaptive quantization without side information using SVQ and TCQ," in Proc. of the 29th Asilomar Conf. on Signals, Sys., and Computers, (Pacific Grove, CA), Oct. 1995.

....of the Laplacian parameter of a given class, based on past quantized data, thus allowing us to further take advantage of localized behavior. The on the fly update of the Laplacian parameter facilitates backward adaptation of the parametric quantizer in a similar way to the ideas introduced in [19] and [20] where non parametric, piecewise linear distributions were used. Finally, we demonstrate how these two techniques can be included in a complete subband image coder using a uniform threshold quantizer (UTQ) 21] as the baseline quantizer. The resulting image coder shows very good ....

....the classification threshold design, based on the unquantized coefficients. We now show how the quantizer can be further refined by estimating k on the fly based on the previously quantized data. Note that the basic idea is extended from the non parametric pdf estimation paradigm introduced in [19] to a parametric case. With the same assumptions and notations from Section II B, consider an L level quantizer with decision levels 0 = b 0 b 1 b 2 : b L 1 for which n j , 1 j L, is the number of data quantized to the j th reproduction level. Note that the quantizer is designed ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," IEEE Trans. on Image Proc., vol. 6, pp. 665--676, May 1997.

....just once. Steinberg and Gutman [20] proposed an algorithm based on string matching with distortion, which achieves R(D=2) for a large class of stationary sources and distortion measures. Zhang and Wei [24] introduced the gold washing method, which sequentially updates the set of codevectors. In [19], the source distribution function is estimated based on the occurrence counts of the scalar quantizer bins, and the quantizer adapts without requiring side information. Our work adopts the one pass approach. Advantages of one pass algorithms include simplicity, computational efficiency, and low ....

A. Ortega and M. Vetterli, "Adaptive quantization without side information," in Proc. ICIP-94, vol. 3, pp. 856--860, 1994.

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