B. Ramamurthi and A. Gersho "Classified vector quantization of images, " IEEE Trans. Commun., vol. COMM-34, pp. 1105--1115, Nov. 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are uniformly lost regardless of the visualization to be performed, the new scheme enables on e to focus more on important features that are more frequently used during the visualization. This approach is different from other enhancement techniques such as the classified vector quantization [ 12] or the classification algorithm [6] that attempt to improve the compression or classification quality based on spatial properties of images themselves. This paper is organized as follows: In Section 2, two popular visualization methods that are considered in this paper are summarized. In Section ....

B. Ramamurthi and A. Gersho. Classified vector quantization of images. IEEE Transactions of Communications, COM34 (11):1105 1115, November 1986.

....A serious problem in ordinary VQ is edge degradation caused by employing the distortion measure such as mean squared error (MSE) To tackle this problem, a classified VQ (CVQ) has been introduced. A variety of CVQ techniques, in which the classification is carried out in the spatial domain [1] or the DCT domain [2] 3] have been proposed in the literatures. However, most developed works were involved with complicated decision making and irregular data flow. Classification in the DCT domain requires complex operations; it also makes the implementation of the classifier difficult. In ....

B. Ramamurthi and A. Gerso, "Classified vector quantization of images", IEEE Trans. Commun., Vol. COM-34, pp. 1105-1115, Nov. 1986.

....a compromise in performance. The fast VQ methods proposed here are applied to each sub vector independently and achieve the same level of performance at about 25 of the full search split VQ complexity. The proposed method is based on classified VQ (CVQ) 4] originally used in image coding [5]. In classified VQ, shown in Figure 1, the input vector is first determined to belong to a certain class out of a predetermined number of classes. Each class is represented by a small set of codevectors, which is a segment ( C in Figure 1) in the optimal codebook. The union of all sets (classes) ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images", IEEE Transaction on Communication, pp.1105-1109, Nov., 1986.

....features is uniformly lost regardless of the visualization to be performed, the new scheme enables one to focus more on important features that are more frequently used during the visualization. This approach is different from other enhancement techniques such as the classified vector quantization [15], the classification algorithm [6] or the multi level techniques [4, 19, 10, 17] that attempt to improve the approximation or classification quality based on spatial properties of images themselves. This paper is organized as follows: In Section 2, two popular visualization methods that are ....

B. Ramamurthi and A. Gersho. Classified vector quantization of images. IEEE Transactions of Communications, COM-34(11):1105--1115, November 1986.

....in the U.K. proposed various improvements to the H.263 scheme. Further important contributions in the field were due to Chen et al. 164] Illgner and Lappe [165] Zhang [166] Ibaraki et al. 167] and Watanabe et al. 168] Vector quantization arrangements were favored by Ramamurthy and Gersho [173] as well as by Torres and Huguet [174] Continuing our overview of nonstandard low rate coding schemes [27] 35] DCT [106] based codecs were proposed in [27] and [32] while various subband coded solutions were advocated by Stedman et al. 125] Woods [126] Gharavi [127] and Ngan [128] ....

B. Ramamurthy and A. Gersho, "Classified vector quantization of images," IEEE Trans. Commun., vol. C-31, pp. 1105--1115, Nov. 1986.

....temporal redundancy, and the variance of the MCER becomes much lower than that of the original image, which ensures bit rate economy. The MCER frame can then be represented using a range of techniques [190] including SBC [144] 145] wavelet coding [146] DCT [80] 81] 188] 191] VQ [149] [151], or QT coding [147] 148] 155] 189] Some 1350 PROCEEDINGS OF THE IEEE, VOL. 86, NO. 7, JULY 1998 Fig. 6. Simple video codec schematic. of these techniques will be highlighted in the forthcoming subsections. When a low codec complexity and low bit rate are required, the motion compensation ....

....a flexible control over the allocation of bits in the spatial frequency domain. The MPEG standard codecs [80] 81] and the H.261 and H.263 codecs scan and entropy code the DCT coefficients and also allow direct encoding of the more correlated video signal on a blockby block basis. VQ [149] [151] can be carried out in both the frequency and time domains, but a persistent deficiency is their difficulty to handle sharp edges adequately. Returning to the DCT principle, our proposed DCTbased codec was designed to achieve a time invariant compression ratio associated with a fixed but ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. Commun., vol. COM-31, no. 11, pp. 1105--1115, Nov. 1986.

....is forced to be assigned to a codeword which will classify it correctly to avoid infinite cost. This, in turn, prevents a codeword from ever changing from its initial class. Thus, for = 1 the BRVQ algorithm reduces to designing a separate codebook for each class. This is essentially classified VQ [7]. However, unlike traditional classified VQ, the class information is not available to the encoder for selection of the appropriate codebook. Instead, the final codebook is simply the union of the separate codebooks. Empirical Study of Performance vs. Every point plotted in this paper ....

....10] If Pr[CX 6= ffi(k) is known or can be estimated from the training set, then I[c X 6= ffi(k) can be replaced with Pr[CX 6= ffi(k) or its estimate in (2) This produces one encoder which can be used both for training and test. For this new encoder, BRVQ with =1 is exactly classified VQ as in [7]. As a result, at =1 it will provide a P e which is exactly the Bayes rule P e if the estimate of Pr[CX 6= ffi(k) is perfect (assuming there are at least as many codewords as classes) For the Diamond and Square problems we saw that BRVQ and OLVQ1 only began to approach the Bayes rule P e for ....

B. Ramamurthi and A. Gersho. Classified vector quantization of images. IEEE Transactions on Communications, COM-34(11):1105--1115, November 1986.

....the range blocks into three distinct classes: shade blocks, midrange blocks and edge blocks. The coder then only checks range blocks in the same class as the current domain block when searching for the optimal transform. Details of the classification algorithm used by Jacquin can be found in [20]. Jacobs, Fisher and Boss [10] classify blocks into 72 different classes and also apply a quadtree partitioning scheme. The quadtree scheme works by using larger blocks (32 Theta 32 in their paper) and splitting the block into four smaller blocks should an error condition not be satisfied. This ....

B. Ramamurthi and A.Gersho. Classified Vector Quantization of Images. IEEE Trans. Commun., 34, Nov 1986.

....the matching primitive is recorded. However, for unsatisfying distortions, the conventional fractal coder is employed to find a better match, after which the primitive dictionary is updated with the new primitive (contracted domain block) and fractal code (affine transformation) Many researches [1, 15, 18, 23, 6, 22, 20] (and many more) have already suggested and established methods to improve the image qualities and search schemes of both coders. However, we restricted ourselves to basic and uncomplicated implementations of the base methods in order to introduce and establish the FVQ coder. This was done because ....

B. Ramamurthi and A. Gersho. Classified vector quantization of images. IEEE Transaction Communications, COM-34, Nov 1986.

....of low volume and rate data suitable for storage in mass memory, and communication over a digital channel. This technique mainly suffers from edge degradation and high computational complexity. Although some more sophisticated vector quantization schemes have been proposed to reduce edge effects ([35]) the computation overhead still exists. Recently, novel approaches have been introduced based on pyramidal structures [1] wavelet transforms [45] and fractal transforms [25] These and some other new techniques [29] inspired by the representation of visual information in the brain, can achieve ....

Ramamurthi, B., Gersho, A., "Classified vector quantization of images", IEEE Transactions on Communications, Vol.COM--34, No.11, pp.1105--1115, November 1986.

....that certain modifications are made. Essentially, this amounts to considering the transformed domains OE(D i ) see (2.3) in place of the original domains. 3.1. 1 Jacquin s approach In his original work [1, 2] Jacquin used a classification scheme coming from a study of Ramamurthi and Gersho [17]. The domain blocks are classified according to their perceptual geometric features. Only three major types of blocks are differentiated: shade blocks, edge blocks, and midrange blocks. In shade blocks the image intensity varies only very little, while in edge blocks a strong change of intensity ....

Ramamurthi, B., Gersho, A., Classified vector quantization of images, IEEE Trans. Commun., COM-34, 1986.

....the optimal subtrees are nested. Henceforth, we will assume that u 1 and u 2 are respectively the average rate and distortion of a finite collection of vector quantizers. 3 Bit Allocation for Classified Vector Quantization Assume that we have a classified vector quantizer (VQ) with M classes [4]. The motivation of classified VQ is to code inputs with codebooks specifically designed for the type of input for better overall performance. The classification can be performed using such methods as decision trees, edge detection, or VQ, and side information is used to specify the class. If ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Transactions on Communications, vol. 34, pp. 1105--1115, Nov. 1986.

....distortion measure in the visual domain, one minimizes the perceptual distortions introduced by the coder. Another class of algorithms attempts to reduce the perceptual distortions by weighting different components in an image based on the response of the human visual system to such components [6, 7, 8]. One example of such a system is described in [6] in which the distortions of blocks with edge patterns are weighted differently from the distortions of blocks with shade patterns or more uniform patterns. A second example weights the distortions in the signal components of the different bands of ....

....the perceptual distortions introduced by the coder. Another class of algorithms attempts to reduce the perceptual distortions by weighting different components in an image based on the response of the human visual system to such components [6, 7, 8] One example of such a system is described in [6] in which the distortions of blocks with edge patterns are weighted differently from the distortions of blocks with shade patterns or more uniform patterns. A second example weights the distortions in the signal components of the different bands of a subband or transform coder on the basis of the ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. Communications, vol. 34, no. 11, pp. 1105-1115, Nov. 1986.

....spectral, perceptual, etc. makes it possible to design quantizers that are better suited for coding the overall signal. In a sense, signal compression with classification is some type of adaptive signal compression. A classified quantizer uses separate codebooks for different classes [1]. Due to the apparent variations of the local statistics of the image blocks, classifying these blocks into more homogeneous groups provides the opportunity for using stationary probabilistic models for each group. Additionally, using a nonuniform bit allocation among the different groups, more ....

B. Ramamurthi and A. Gersho, "Classified Vector Quantization of Images," IEEE Trans. Commun., pp. 1105--1115, Nov. 1986.

....frequency of 1800 MHz and a pedestrian speed of 2 mph. 1 Introduction Previously proposed fractal codec designs were targeted at high resolution images having large intra frame domain block pools [1] 2] Following the approaches proposed by Barnsley [3] Jacquin [2] Monro [1] 5] Ramamurthi [6] et.al and Beamont [4] in this study we explored the range of tradeoffs available using five different head and shoulders fractal video phone codecs (Codecs A E) and compared their complexity, compression ratio and image quality specifically for low resolution, small pool 176 Theta 144 pixels ....

....gradient, 26.39 edge detected (total) 0 deg 2.66 45 deg 1.85 90 deg 5.69 135 deg 3.18 180 deg 3.92 225 deg 2.96 270 deg 3.92 315 deg 2.22 Mixed Edge angle ambiguous 9. 24 Table 2: Classified Block Types and Their Relative Frequencies in Codec E improved by a block classification algorithm [2] [6]. Accordingly, the image blocks were classified into four classes: 1. Shade blocks taken from smooth areas of an image with no significant gradient. 2. Midrange blocks having a moderate gradient but no significant, edge. 3. Edge blocks having steep gradient and containing only one edge. 4. Mixed ....

B. Ramamurthi and A. Gersho, "Classified Vector Quantization of Images", IEEE Trans. Commun., Vol 34 No 11, No 86, pp 1258-1268

....and classification can also be obtained using a sequential classifier quantizer design, in which classification is performed first, and then quantization is performed on the results of the classifier. One approach is to design separate quantizers for each class, yielding a classified VQ [27]. This technique has been investigated primarily for the sole objective of improved compression, although Gorman [1] and Wesel et al. 28] have examined such a method for the joint goal. Another is to use a classification scheme such as Stone s generalized nearest neighbors [29] clustered ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images", IEEE Trans. Comm., vol. COM-34, no. 11, pp. 1105--1115, Nov. 1986.

....DCT bases was used to select the best basis. The coding was performed in a JPEG like environment using a MSE distortion criterion. In order to have the quantizers adapt to the signal s non stationarities, classified quantizers were chosen as the admissible set, similar to the VQ application of [21]. Four quantizer classes were used, optimized for 1) typical image blocks with low frequencies weighted much higher than the perceptually less sensitive higher frequencies, like the JPEG suggested matrix; 2) horizontal edges; 3) vertical edges; and (4) image blocks with a white frequency ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. on Commun., vol. 34, pp. 1105--1115, Nov. 1986.

....on generalized product code VQ (GPC VQ) where a source is sequentially coded in stages and the codebook for each stage depends on the outcome of the previous stage [5] The use of constrainedstorage methods is essential for GPC VQ. As a final example, we observe that in classified VQ (CVQ) 1] [6] and finite state VQ (FSVQ) 7] 20] the critical performance limitation is the storage of a large number of class state specific codebooks. B. Relevant Prior Work An earlier approach to mitigating the problem of storage complexity in VQ, called constrained storage VQ (CSVQ) was introduced by ....

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. Commun., vol. COM-34, pp. 1105--1115, Nov., 1986.

No context found.

B. Ramamurthi and A. Gersho "Classified vector quantization of images, " IEEE Trans. Commun., vol. COMM-34, pp. 1105--1115, Nov. 1986.

No context found.

B. Ramamurthi, A. Gersho, "Classified vector quantization of images", IEEE Transactions on Communications, vol COM-34, november 1986

No context found.

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. Commun., vol. COM-28, pp. 84-95, Jan. 1980. 19

No context found.

B. Ramamurthi and A. Gersho, Classified Vector Quantization of Images, IEEE Trans. Commun. COM-34(11), pp11051115, Nov, 1986.

No context found.

B. Ramamurthi and A. Gersho, "Classified Vector Quantization of Images," IEEE Trans. on Com., pp. 1105--1115, Nov. 1986.

No context found.

B. Ramamurthi and A. Gersho, "Classified vector quantization of images," IEEE Trans. Commun., Vol. COM-34, pp. 1105-1115, Nov. 1986.

No context found.

B. Ramamurthi and A. Gersho, `Classified vector quantization of images', IEEE Trans. Comm. Vol. COM-34 No. 11 pp. 1105-1115, 1986.

First 50 documents

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