Video delivery over wireless and cellular networks can be severely impaired due to bandwidth limitations and also due to the presence of burst losses and signal-loss intervals. This paper proposes unequal protection of Region of interest (ROI) encoded videos for transmission over low-bitrate and error-prone channels. Here, a flexible and interactive ROI is introduced in a low-complexity, and standard-compliant fashion to encode the critical regions within each frame at a higher quality as compared to the background. Further, the reconstruction fidelity of the high-quality ROI is improved by unequally protecting the ROI using an error-resilient video delivery scheme known as Multiple Representation Coding (MRC). Simulation results indicate that the proposed strategy can facilitate a graceful recovery of the ROI in the presence of burst and signal losses.

Heterogeneity of the Internet and clients' operating environment necessitate the adaptation of media content to ensure a satisfactory multimedia experience. Scalable video coding (SVC) is a layer-based encoding scheme that achieves this adaptation through the extraction of partial bitstreams at reduced spatial, quality and temporal resolutions. The main focus of this paper is in developing such an extraction algorithm in the context of video streaming applications. Based on a combination of metadata computations and prediction mechanisms, our algorithm evaluates the quality contribution of each layer in the SVC bitstream and makes extraction decisions that are aimed at maximizing video quality while operating within the available bandwidth resources. When compared to earlier extraction methods such as JSVM-Basic and JSVM-QL, our algorithm offers a quality gain of 1.6 dB and 0.5 dB respectively. The proposed technique also achieves a 73% reduction in metadata computation time when compared to JSVM-QL.

The papers in this collection are diverse, and address multiple application genres: entertainment, collaboration, surveillance, and healthcare, all powered by new combinations of enablers - scientific, technological, and standards driven. The expositions are generally too multifaceted to permit simple categorization. However, we have arranged these contributions in a sequence that the reader may find useful.

Delivering affordable high-quality E-health care services requires effective telecommunication and multimedia technologies in conjunction with medical expertise. In this paper, we propose and demonstrate a telecommunication and networking architecture for implementing next generation Telemedicine and Telehealth systems. Our integrated optical-wireless based network provides super broadband, ultra low-latency connectivity for voice, video, image and data across various telemedicine modalities to facilitate real-time and near real-time communication of remote health care information.

Although small screen sizes and limited input methods are challenges for mobile devices, there are affordances like gestures and motion detection through on-device sensors. A prototype system has been implemented on a HTC Dream (T-Mobile G1) mobile phone demonstrating an "adaptive TV-channel" like mixed-initiative interface used to solicit user relevance feedback, provide recommendations and facilitate news video watching. The user is provided on-screen finger gesture operations to vote-up, vote-down or skip a video. Shaking the device resets the video sequence. This creates a cognitively palatable stream of videos and a seamless lower-latency user experience easily operated with one hand.

This paper evaluates the performance of a zero reference metric used in objective measurement of the quality of video which has been transcoded at various bit rates to suit the needs of the 3G mobile environment. This Zero reference metric (AVQ) has been shown to exhibit a high correlation with Mean Time Between Failures (MTBF) which is a functional measure of video quality representing how often a typical viewer observes a noticeable visual error. This paper reports on the performance of transcoders with MPEG2 video at 512 kbps as input and 192, 256 and 384 kbps versions of MPEG-4 Part 2 Visual and MPEG-4 Part 10 (H.264 ) as outputs. We use AVQ-estimated MTBF and PSNR as quality metrics. Our results show that MTBF better reflects subtleties such as lowered visual quality during significant motion. It also provides a more descriptive spread of quality in the 192-384 kbps range of transcoder outputs.

As digital communication of television content becomes more pervasive, the long-standing problem of assessing video quality becomes particularly important. This work describes some innovative guidelines for easy and reliable determination of a quality metric that is subjectively meaningful: mean time between failures (MTBF), representing how often a typical viewer observes a noticeable visual error and a related instantaneous metric relating to the fraction of viewers that find a given video portion to be within acceptable quality levels. The value of MTBF is addressed in the context of video quality, and objective measurements that correlate well with subjective evaluations of MTBF are investigated for different video clips at bit rates in the range of 1.5 - 5 Mbps. The full-reference objective metrics of PSNR and JND are found to have correlation coefficients of 0.69 and 0.88 respectively. In contrast, a reduced-reference objective metric, spatial temporal join metric (S. Wolf and M.H. Pinson, 1999) (STJM) is found to have a correlation coefficient of 0.84. A method for estimating the MTBF of a video sequence from objective measurements is also described

Wireless communication of video poses constraints on information capacity. Region-of-interest (ROI) video coding provides higher quality in the ROI, but poorer quality in the background (BKGRND), for a given total bitrate (TBR). Researchers, including the authors, have also proposed more graceful quality management methods, using what is referred to here as an extended-region-of-interest (EROI). We consider three levels of losslessness-mathematical, diagnostic, and perceptual, with the goal of associating them with the above-mentioned regions. We describe work in progress aimed at optimizing an elastic expert system based on the above methodology, with telehealth video as its anchor. The optimizations are threefold-user, perceptual, and network oriented, and are incorporated in the rate control algorithm. We propose a rate control method where, unlike conventional methods, bit allocation is shifted from the frame level to individual regions within the frame. Thereafter, the above-mentioned criteria are used to determine regional bit allocation. Peak-signal-to-noise-ratio (PSNR) results show, as expected, that the proposed scheme achieves higher ROI-EROI quality than the verification model VM8 of MPEG4. This is illustrated with four examples of pediatrics video. The value and design of the proposed methodology is being corroborated by subjective testing involving medical experts. We are independently researching another standing issue in the telehealth application, that of low complexity segmentation and tracking of the ROI-EROI boundaries

A system is being created with the aim of providing news video to mobile devices, focusing on using an "interactive TV-channel" metaphor as one of the possible interaction modes. The user is presented a sequence of recommended news video clips and a set of operations to vote-up or vote-down a video to indicate its relevance. User-machine synergy is harnessed through interaction with the clips, used to tailor subsequent media clip recommendations. A framework for delivering relevant news clips to a mobile device is being developed, and future work and challenges are discussed in further detail. The goal is to create a methodology for presenting a sequence of news video clips in a manner that is seamless and cognitively palatable through usermachine synergy, and switching between automated and manual modes as deemed necessary.

Three-dimensional (3-D) ray-tracing and analysis is used to characterize the path angles in a rectangular building in which all interior walls are parallel or perpendicular to the exterior walls. Path angles are recorded for 500 random transmitter and receiver locations, maintaining constant range throughout a particular building model. The main contribution of this paper is the observation that when all of the path angles from all 500 trials are measured relative to one building wall and collectively analyzed in a signal strength distribution, clustering is clearly apparent. While this angle clustering in the ensemble of channels is not the same as clustering in a single channel trial, it may partially account for the clusters observed in single trials. As a practical matter, these results may impact the use and placement of directional antennas for wireless LANs in buildings of this type.

We describe an efficient bit allocation algorithm using 2D interpolation of the rate-distortion (R-D) characteristics in a dependent coding environment for applications of one-way video. To solve the dependent coding problem, it is necessary to measure R-D data for all possible combination of quantization settings in the reference frame and the predictive frame. 2D interpolation of rate-distortion characteristics is performed with only a few R-D control points to reduce complexity without significant loss of performance. We also explain a pruning algorithm to further reduce the computational complexity in obtaining R-D characteristics. The cost function for each quantization pair is obtained using 2D R-D characteristics. The complexity is reduced by the 2D-interpolation and pruning algorithm without significant loss of optimality. Experiments with MPEG-2 show that the proposed algorithm provides improvement of PSNR by 0.9/spl sim/1.8 dB over the MPEG-2 TM5 algorithm.

Summary form only given. With the emergence of broadband telecommunications, we are moving in well-measured steps from nearly universal telephony to increasingly pervasive multimedia services. The author describes some of the fundamental opportunities that this context presents to the practitioners of advanced signal processing. These opportunities span conventional media processing disciplines such as signal capture, compression, synthesis, recognition and information retrieval, as well as a rich collection of technologies that has come to be known as communications signal processing. He remarks particularly on opportunities for enhancing lifestyle in the broadband-enabled home of the future.

There are several papers published in the past few years about the evaluation of IEEE 802.11(b) WLAN performance degradation due to Bluetooth radio interference. The network topologies used WLAN in the previous literature are point-to-point, i.e., on AP and one WLAN station. In this paper we evaluate a performance of IEEE 802.11(b) systems composed of an AP and multiple WLAN stations using the same 2.4 GHz frequency band in the presence of Bluetooth radio systems within a BSS. Our results focus on the impact of Bluetooth interference on IEEE 802.11(b) WLAN systems. the throughput of IEEE 802.11(b) is estimated as a function of not only WLAN parameters but also Bluetooth radio system parameters. Further, we derived theoretical maximum throughputs of IEEE 802.1(b) systems with perfect channel conditions in order to compare with other scenarios that considering interferences and competition with other WLAN stations. As a result, with light and heavy BT usage scenarios, the IEEE 802.1(b) system throughputs are degraded by 25% and 66% respectively.

In this paper, we describe an adaptive non-linear motion vector resampling algorithm for video transcoding where spatial resolution down-scaling is involved. The incoming MPEG-2 video is spatially down-scaled by a factor of two, and transcoded to MPEG-1 video at lower bit rate. Since the heaviest computational load of encoding is the motion estimation process, motion vectors of the original video have to be reused from the bit stream. However, reusing motion vectors with down-scaling is not easy, since the four motion vectors to be down-scaled from original video are usually not well aligned. The motion vector reusing scheme greatly affects the performance of the transcoder since the decoded video quality depends on the performance of motion compensation as well as rate control. The performance of an alternative weighted averaging scheme tends to be degraded by inappropriate motion vectors. Experiments show that a motion vector selection algorithm indeed performs better than a weighted average method. The selection algorithm uses a likelihood score based on the investigation of statistical characteristics of the macroblocks associated with the best matching motion vectors. The motion vector having the highest likelihood score is selected to be reused. The weighted averaging scheme is also applied if there are macroblocks having the same highest scores. The experimental results with various bit rate and sequences show that the proposed algorithm provides improvement of PSNR by 0.2/spl I.utilde/0.7 dB on average and that the computational complexity is not much higher. Therefore, the overall transcoding performance and efficiency can be increased by the proposed algorithm.

We present an information fusion approach to designing a semantics-sensitive image retrieval system. Our approach is based on the three different types of classifiers, which extract and provide semantic cues of image regions. This local information from the proposed classifiers is fused together to generate the semantic labels of images. The experimental results show that our information fusion approach makes image retrieval more semantics-sensitive and improves the image retrieval performance significantly.

This work describes the optimization of a signal processing front-end for a distributed speech recognition system with the goal of reducing power consumption. Two categories of source code optimizations were used, architectural and algorithmic. Architectural optimizations reduce the power consumption for a particular system, in this case, the HP Labs Smartbadge IV prototype portable system. Algorithmic optimizations are more general and involve changes in the algorithmic implementation of the source code to run faster and consume less power. A cycle accurate energy simulation shows a reduction in power usage by 83.5% with these optimizations. The optimized source code runs 34 times faster than the original code, therefore it can run at lower processor clock speeds and voltages for further reductions in power consumption. This technique, known as dynamic voltage scaling, was implemented on the Smartbadge IV hardware for an overall reduction in power usage of 89.2%.

It was recently shown that the full service access network (FSAN) method for computing the power spectral density (PSD) of the crosstalk noise produced by a mixture of digital subscriber line (DSL) disturbers may be considered a particular case of a general lower bound-obtained using Minkowski's inequality-of the sum of the power spectra of the noises generated by the individual interferer classes. Such a result was deemed important because it gave certain mathematical validation to the FSAN method, which currently lacks a theoretical foundation. The main contribution of the present paper is a proof that the FSAN method is equivalent to the L/sup P/ norm of a function suitably defined in a finite counting measure space, for a particular value of p. This result is regarded as relevant to the search of a mathematical basis for the FSAN method because it brings forward a new perspective for the theoretical analysis of the technique. Several results obtained with the help of this formulation are presented, including (1) a proof that the FSAN method always produces results smaller than or equal to the ones obtained by the simple sum of the crosstalk PSDs of the individual interferer classes and (2) a proof that the L/sup P/ norm with p/spl rarr//spl infin/ may be used to estimate the PSD of the crosstalk produced by the dominant disturber or disturber class at a particular frequency. The meaning of these results is illustrated with two examples.

We present a new definition of data hiding capacity which complements the established theory in the field and produces practical estimates under many attacks. We discuss the relation between the proposed definition and the current theoretical work on data hiding capacity. The definition proposed is applied to still images to estimate the hiding capacity of a particular image under attacks such as JPEG compression and additive noise.

This paper presents an adaptive forward error correction (AFEC) protocol that provides a reliable communication service for real-time traffic over low-earth orbit (LEO) satellite networks. In time-varying wireless links, such as LEO satellite networks. A reliable channel estimation scheme with an appropriate code selection technique is essential for adaptive error-control systems. This paper proposes a new channel estimation scheme that uses receiver-initiated messages (ACKs, NAKs, and INCs). These messages feed back to the transmitter which then selects the code rate for sending packets. In addition, the scheme uses the concept of a dynamic transmittable code set by adjusting the maximum code rate in the set of possible codes from which the transmitter can select. In terms of throughput (from network provider's viewpoint) and packet error rate (from subscriber's viewpoint), performance results show that the proposed scheme guarantees the quality of service (QoS) requirements of real-time applications.

Content based image retrieval is one of the most active research areas in the field of multimedia technology. Currently, the relevance feedback approach has attracted great attention since it can bridge the gap between low-level features and the semantics of images. We propose a new relevance feedback technique, which uses the normal mixture model for the high-level similarity metric of the user's intention and estimates the unknown parameters from the user's feedback. Our approach is based on a novel hybrid algorithm where the criterion for the selection of the display image set is evolved from the most informative to the most probable as the retrieval process progresses. Experiments on the Corel image set show that the proposed algorithm outperforms MindReader at the semantics based search.

Clustering of paths in angle in RF multipath scenarios has been observed in measured data, explicitly used in stochastic propagation models, and shown to have a significant impact on the capacity of MIMO links. A 3D ray tracing is used to characterize the path delays and angles in a rectangular building in which all interior walls are parallel and perpendicular to the exterior walls. Path delays and angles are recorded for 200 random transmitter and receiver locations, maintaining constant range and throughout a particular building model. The main contribution of this paper is the observation that when all of the path angles from all 200 trials are measured relative to one building wall and collectively analyzed in a histogram, clustering is clearly apparent. Furthermore, the mean angles of the clusters are a function of the range between the transmitter and receiver. While this angle clustering in the ensemble of channels is not the same as clustering in a single channel trial, it may partially account for the clusters observed in single trials. As a practical matter, these results may impact the use and placement of directional antennas for wireless LANs in buildings of this type. By simply measuring the angles relative to the line-of-sight (LOS) path rather than the building wall, the data is shown to be consistent with the joint probability density function (PDF) of delay and angle based on the geometrically based elliptical model (GBEM).

We describe a source and channel coding algorithm suitable for video messaging over wireless networks. The research platform is based on ITU-T H.263 video coding, Reed Solomon channel coding, data interleaving and a wireless network provided by Lucent's WaveLAN. We consider the special characteristics of one-way video messaging, and propose rate control, unequal error protection and error concealment algorithms with features for object-oriented quality enhancement. These features consist of: (1) object-oriented bit allocation based on the segmentation of video into objects of interest, (2) unequal channel error protection by object-oriented data partitioning, and (3) object-oriented error concealment using the locations and motion vectors of object. Based on experiments with head and shoulders scenes, the subjective quality and error resilience of decoded video are seen to increase significantly. Simulation results show that the proposed algorithm is effective at high packet loss rates and with clustered packet losses.

This paper presents a novel postprocessing algorithm developed specifically for very low bit-rate MC-DCT video coders operating at low spatial resolution, postprocessing is intricate in this situation because the low sampling rate (as compared to the image feature size) makes it very easy to overfilter, producing excessive blurring. The proposed algorithm uses pixel-by-pixel processing to identify and reduce both blocking artifacts and mosquito noise while attempting to preserve the sharpness and naturalness of the reconstructed video signal and minimize the system complexity. Experimental results show that the algorithm successfully reduces artifacts in a 16 kb/s scene-adaptive coder for video signals sampled at 80 x 112 pixels per frame and 5-10 frames/s. Furthermore, the portability of the proposed algorithm to other block-DCT based compression systems is shown by applying it, without modification, to successfully post-process a JPEG-compressed image.

Voice and gesture represent fundamental and universal modalities in interhuman communication. With recent advances in automatic methods of speech recognition and synthesis, human-machine interaction by voice is rapidly becoming a technological and commercial reality. Although less mature and deployed, gesture recognition by machine is becoming reliable enough to be considered as a serious supplement to the voice interface between humans and machines.

This paper proposes a new normalized parametric rate-distortion model which can be efficiently applied to H.263-compatible video codecs and other codecs of similar type. The model is very efficient in terms of computational complexity and amount of required memory since it approximates all 31 rate-distortion relations (one for each quantization parameter QP) with only one curve by normalizing the energy of signals and the quantization distortion with functions of QP. We show how the model parameters can be obtained from experimental data. We also propose a quantizer selection scheme which uses the model to estimate the number of bits required for encoding 8/spl times/8 residual blocks. Experiments show that the model provides very good approximations of the rate and the distortion characteristics for all QPs, and that the proposed quantizer selection scheme works better than the scheme used in the MPEG-4 Video Verification Model Version 5.1.

A new adaptive postprocessing algorithm to enhance the quality of a noisy video sequence is presented. The algorithm recognizes that the visibility of noise depends on local signal characteristics. It therefore classifies the video signal into different classes and uses separate nonlinear filters matched to each class. The most general version of the algorithm employs motion-compensated frame averaging to improve picture quality in a first stage. A classification algorithm subsequently divides subblocks of pixels in the averaged frame into four classes: edge, smooth, nonsmooth with motion and nonsmooth without motion. Spatial algorithms that perform multilevel median filtering, double median filtering, and median filtering are used for pixels belonging to edge, smooth, and nonsmooth with motion categories. Pixels in the nonsmooth, unmoving category are left unfiltered to preserve corresponding image texture. In a simpler version of this four-class system, the motion cues and motion-compensated frame averaging are eliminated, and the purely spatial filtering is based on a three-class algorithm. When used at the output of a 3-D subband coder at 384 kbps, the spatial postfilter was shown to provide a consistent gain in subjectively evaluated picture quality. Twenty-five viewers participated in an experiment involving three coded sequences. In a pairwise comparison of postfiltered and unfiltered sequences, the postfiltered version was judged to be better in 63 out of 75 instances.

We describe and show the results of video coding based on a three-dimensional (3-D) spatio-temporal subband decomposition. The results include a 1-Mbps coder based on a new adaptive differential pulse code modulation scheme (ADPCM) and adaptive bit allocation. This rate is useful for video storage on CD-ROM. Coding results are also shown for a 384-kbps rate that are based on ADPCM for the lowest frequency band and a new form of vector quantization (geometric vector quantization (GVQ)) for the data in the higher frequency bands. GVQ takes advantage of the inherent structure and sparseness of the data in the higher bands. Results are also shown for a 128-kbps coder that is based on an unbalanced tree-structured vector quantizer (UTSVQ) for the lowest frequency band and GVQ for the higher frequency bands. The results are competitive with traditional video coding techniques and provide the motivation for investigating the 3-D subband framework for different coding schemes and various applications.

A new adaptive post-processing algorithm to enhance the quality of a noisy video sequence is presented. This algorithm employs motion compensated frame averaging to improve picture quality. A classification algorithm divides sub-blocks of pixels in the averaged frame into 4 classes: edge, smooth, non-smooth with motion and non-smooth without motion. Spatial algorithms that perform multilevel median filtering, double median filtering and median filtering are used for pixels belonging to edge, smooth and non-smooth with motion categories. Pixels in the nonsmooth, unmoving category are left unfiltered to preserve corresponding image texture. In a simpler version of this 4-class system, the motion cues and motion compensated frame averaging are eliminated, and the purely spatial filtering is based on a 3-class algorithm. When used at the output of a 3-D subband coder at 384 kbps, the spatial postfilter was shown to provide a consistent gain in subjectively evaluated picture quality. Twenty-five viewers participated in an experiment involving three coded sequences. In a pairwise comparison of postfiltered and unfiltered sequences, the postfiltered version was judged to be better in 63 out of 75 instances.>

The notion of perceptual coding, which is based on the concept of distortion masking by the signal being compressed, is developed. Progress in this field as a result of advances in classical coding theory, modeling of human perception, and digital signal processing, is described. It is proposed that fundamental limits in the science can be expressed by the semiquantitative concepts of perceptual entropy and the perceptual distortion-rate function, and current compression technology is examined in that framework. Problems and future research directions are summarized.>

The CCITT is in the process of standardizing a 16 kb/s speech coder submitted by AT&T called low-delay code excited linear prediction (LD-CELP). In the first phase of CCITT testing, the coder met all performance requirements except for the tandeming condition. To improve the tandeming performance, the authors first tuned the perceptual weighting filter to optimize the coder's performance after three tandems. They then added an adaptive postfilter which was also tuned for three tandems. The excitation codebook was also reoptimized using a multiple-language IRS-weighted training database. With these changes, in the second phase of CCITT testing, the coder significantly exceeded the tandeming performance requirement and also met all other requirements. In fact, the speech quality of LD-CELP was equivalent to or better than that of 32 kb/s ADPCM for all conditions tested. It is anticipated that the version of LD-CELP will be formally ratified as a CCITT standard in 1992. The author describes the modifications made to improve the performance and discusses the CCITT test results.>

A low-delay code-excited linear prediction (LD-CELP) speech coder which is expected to be standardized in 1992 as a CCITT G Series Recommendation for universal applications of speech coding at 16 kb/s is presented. The coder achieves a one-way coding delay of less than 2 ms by making both the LPC predictor and the excitation gain backward-adaptive and by using a small excitation vector size of five samples. The official CCITT laboratory tests revealed that the speech quality of this 16 kb/s LD-CELP coder is either equivalent to or better than that of the CCITT G.721 standard 32-kb/s ADPCM coder for almost all conditions tested. A description of the LD-CELP algorithm, its implementation on the DSP32C for CCITT testing, and performance results from these tests are presented.>

A description of technology targets in signal compression and a nonexhaustive account of research directions that may lead toward these targets are presented. Opportunities for integrating source coding and channel coding technologies are also pointed out. Such integration, which has hitherto been an informal exercise, will become increasingly essential as communication capabilities are stretched with capacity-limited channels such as wireless media. In parallel, as greater sophistication is sought in the integration of speech and data with broadband signals such as CD-audio and high-resolution video, there will be increased interaction of signal compression technology with the field of communication networking.>

A full-motion video coder based on a threedimensional subband framework is developed. The original image data is decomposed into different spatial-temporal frequency bands and based on the characteristics of the data in each band, different quantization strategies are applied to each band. The lowest spatial-temporal frequency band is coded using a three dimensional switched predictor and optimum scalar quantizer. The nondominant subbands which contain perceptually significant data are coded using a vector quantization approach in which the sparse, highly structured nature of the data in the higher-frequency subbands is exploited in the design of the codebook.>

The authors explore the benefits of time-varying bit allocation to excitation and LPC (linear predictive coding) parameters for the case of codebook-excited LPC. The overall bit rate in the experiment was 4.8, 6.4, or 8.0 kb/s. In each case, permissible bit rates for the LPC component were 0, 24, 36, or 48 bits per frame, one of which was selected for each speech frame using a brute-force search maximum performance. Average SNR gains over conventional time-invariant methods were modest, on the order of 1 to 2 dB, but gains for certain speech segments were as high as 3 to 5 dB. Perceptually, gains due to variable bit allocation were most noticeable in the 6.4 kb/s system, especially with female speakers. However, even in this case, the benefits of flexible bit allocation were somewhat offset by distortions due to other inadequacies in the coding algorithm.>

A generalization of the K-means clustering algorithm to include spatial constraints and to account for local intensity variations in the image is proposed. Spatial constraints are included by the use of a Gibbs random field model. Local intensity variations are accounted for in an iterative procedure involving averaging over a sliding window whose size decreases as the algorithm progresses. Results with an eight-neighbor Gibbs random field model applied to pictures of industrial objects and a variety of other images show that the algorithm performs better than the K-means algorithm and its nonadaptive extensions.>

A class of image quantizers called model-testing vector quantizers is introduced. These are finite-state quantizers with the additional ability to make a decision about whether the current input conforms to the assumed finite-state model or not. The nonconforming component is coded at relatively lower bit rates by utilizing well-known linear and nonlinear properties of the human visual system. An adaptive spatial and intensity quantization procedure is used that removes the perceptually irrelevant information from the nonconforming component. The differential brightness discrimination of the eye is also utilized. Resulting coded images are indistinguishable from the originals at bit rates that range from 1.0 to 3.5 bits/pixel for a wide variety of 512*512 images. These bit rates can be regarded as upper bounds for the perceptual entropy of the corresponding images.>

Multistage quantization is used to encode the intensity waveform in the most dominant subband. The reconstruction error in the first stage of the process is itself quantized and first-stage reconstruction is refined by removing from it the most objectionable components of quantized reconstruction error. The reconstruction error is coded on the basis of a bit allocation algorithm that utilizes the fact that perceptually significant error samples tend to occur in clusters in high-entropy parts of the input image frame. Application of this technique to a 512*512 monochrome image results in good communications quality at a rate of 0.5 bit per pixel for the simplest case of a two-band partition in horizontal and vertical frequencies, for a total of four subbands. By increasing the number of subbands to 16, significantly higher image quality can be realized at the same bit rate.>

A generalization of the K-means clustering algorithm to include spatial constraints and to account for local intensity variations in the image is proposed. Spatial constraints are included by the use of a Gibbs random field model. Local intensity variations are accounted for in an iterative procedure involving averaging over a sliding window whose size decreases as the algorithm progresses. Results with an eight-neighbor Gibbs random field model applied to pictures of industrial objects and a variety of other images show that the algorithm performs better than the K-means algorithm and its nonadaptive extensions. >

Two very different subband coders are described. The first is a modified dynamic bit-allocation-subband coder (D-SBC) designed for variable rate coding situations and easily adaptable to noisy channel environments. It can operate at rates as low as 12 kb/s and still give good quality speech. The second coder is a 16-kb/s waveform coder, based on a combination of subband coding and vector quantization (VQ-SBC). The key feature of this coder is its short coding delay, which makes it suitable for real-time communication networks. The speech quality of both coders has been enhanced by adaptive postfiltering. The coders have been implemented on a single AT&T DSP32 signal processor.>

It is shown that postfiltering circuits based on higher order LPC (linear predictive coding) models can provide very low distortion in terms of special tilt. Thus, they can provide better speech enhancement than circuits based on the backward-adaptive pole-zero predictor in ADPCM (adaptive digital pulse code modulation). Quantitative criteria for designing postfiltering circuits based on higher-order LPC models are discussed. These postfilters are particularly attractive for systems where high-order LPC analysis is an integral part of the coding algorithm. In a subjective test that used a computer-simulated version of these circuits, enhanced ADPCM obtained a mean opinion score of 3.6 at 16 kb/s.>

The quality of low bit rate ADPCM speech can be enhanced by using a noise-shaping postfilter whose parameters can be derived from those of the ADPCM predictor. The speech distortion that is inevitable in postfiltering can be mitigated by adapting the degree of postfiltering to ADPCM performance. In order to reduce speech distortion to even lower values while maintaining adequate noise reduction at 2 bits/sample, we propose an arrangement where the noise shaping is realized in two stages. The first stage, an all-pole noise shaper, is realized at the encoder by means of reconstruction noise feedback. The second stage, which is essentially an all-zero noise-shaping circuit is realized using a postfilter. The distortion of the speech is now confined to an all-zero operation, and is much less perceptible than with a pole-zero postfilter. The parameters of the noise shaping and postfiltering circuits are obtained very simply from corresponding parameters of a backward-adaptive pole-zero predictor. The result of the above arrangement is 2-bit ADPCM coding with zero delay and very high communications quality. In a subjective test that included several real-time coders at 16kb/s, the proposed ADPCM algorithm scored a mean opinion score (MOS) of nearly 3.5 on a scale of 1 to 5. This compares with an MOS value of under 3.0 for ADPCM without noise shaping or postfiltering, and MOS values in the range of 3.7 to 3.8 for sub-band and multipulse coders.

Adaptive differential PCM (ADPCM) is known to provide high quality digitization of telephone bandwidth speech at 32 kb/s. For ADPCM systems operating at sub-standard rates such as 24 or 16 kb/s, a postfiltering technique has been shown to provide a simple means for enhancing speech quality [1]. This paper proposes an algorithm which adapts the degree of postfiltering to the local performance of the ADPCM coder. The adaptation requires no extra information from the transmitter. As a result of the adaptation, segments of speech which are reproduced relatively well by the coder are only mildly postfiltered. At 16 kb/s, the adaptively postfiltered system provides good communications quality with most telephone speech inputs.

In this paper we propose a hybrid speech coder which utilizes properties of APC, ATC, and subband coding. Subband splitting is used to reduce the dynamic range of a full-band APC. As a result, the prediction or power gain is reduced and the instability problem associated with the APC coder is alleviated to a large extent. The optimal bit allocations used in ATC is extended in the new coder where the available coding bits are optimally allocated, both in the time and frequency domains. Frame boundary artifacts (such as those found in ATC) are not present due to the time-domain processing nature of SBC. In order to improve the efficiency in quantizing the side information (time and frequency domain signal power and the corresponding bit allocations), vector quantization is used. Adaptive backward predictors are used to further reduce the number of bits allocated to side information, leaving more bits to encode the prediction residual signals. At 16 kbps the coder achieves better than 20 dB segmental signal-to-noise ratio and sounds transparent for most speakers.

In this study it is our goal to improve the performance of ADPCM and subband speech coders at medium bit rates (9.6∼16 kb/s) without increasing the coder complexity substantially. Various major building blocks including the predictors (both fixed and adaptive), subband quadrature mirror filter (QMF) and bit assignment strategy (both static and dynamic) are investigated in detail. We have found that (1) the Least-Squares (LS) adaptive lattice predictor outperforms both the pole-zero adaptive predictor recommended by CCITT and a first-order fixed predictor. (2) more subbands can improve the coder performance (3) longer QMF can reduce the interband aliasing and improve the subjective performance of a subband coder (4) an optimal dynamic bit allocation scheme with an improvement of SNR as high as 5 dB is much more favorable than a fixed bit allocation. With all of the above finding we propose a 4-band hybrid subband coder with an LS adaptive lattice predictor and an optimal dynamic bit allocation strategy.

A first-order Markov process is used to model the sequence of quantization noise samples in delta modulation. An autocorrelation parameterCin the Markov model controls the shape of the noise spectrum, and asCdecreases from 1 to 0 and then to -1, the spectrum changes from a low-pass to a flat, and then to a high-pass characteristic. One can also use the Markov model to predict the so-called out-of-band noise rejection that is obtained when delta modulation is performed with an oversampled input, and the resulting quantization noise is lowpass filtered to the input band. The noise rejectionGis a function ofCas well as an oversampling factorFand an interesting asymptotic result is thatG=\frac{1-C}{1+C} \dot FifF \gg \frac{1+C}{1-C} \dot \frac{\pi}{2}. Delta modulation literature has noted the importance of the specialGvalues,Fand2F. These correspond to autocorrelation values of 0 and -1/3.

This paper shows the utility of using adaptive quantizers in the tree-encoding of speech waveforms based on the (M, L) algorithm [1]. Resulting adaptive differential PCM (ADPCM) and adaptive delta modulation (ADM) encoders, with time-invariant prediction networks, can provide useful speech outputs at bit rates in the order of 24 kbits/s; at 16 kbits/s, on the other hand, the encoders exhibit clearly perceptible amounts of quantization noise.

In digital speech communication, transmission errors generally introduce impulsive distortions in the received speech waveform. Smoothing of this waveform results at once in a squelching of the distortion component, and in an undesirable smearing of the speech. However, our experience with practical differential PCM (DPCM) codes (with adaptive quantizers and first-order predictors) has shown that if the error probability is fairly significant (for example, 0.025), the noise attenuation is perceptually desirable in spite of the attendant speech-muffling. Our observations are based on computer simulations and informal listening tests. The smoothing can be performed either on the received DPCM word (prediction error signal) or the reconstructed speech amplitude. (The two signals are identical in nondifferential PCM.) Smoothing algorithms can be either linear (based, for example, on running averages) or nonlinear (based, for example, on running medians). Studies with 3-bit quantizers indicate that with independently occurring transmission errors, smoothing of the prediction error signal is perceptually desirable, although the benefits decrease as a function of the predictor coefficient a, with the maximum advantage showing up fora = 0(PCM). Running averages and running medians seem to work equally well, and suggested block lengths for their computations are three or five samples. Results with clustered transmission errors show a higher advantage due to smoothing and a preference of linear methods and longer block lengths.