We propose a joint source-channel modeling approach for adaptively determining the quality-optimal FEC code rate. Our objective is to obtain the optimal video quality in the receiver, while taking time-varying packet loss into consideration. First, we define the loss threshold set as the set of packet loss probabilities in which the code rate have to be adjusted to maintain the maximum video quality. Proposed model can estimate the loss threshold set efficiently. Then, whenever there is a change of the packet loss status in the channel, video sender can always find the optimal code rate on-the-fly by using the estimated loss threshold set. Simulation results show that the proposed method can determine the near-optimal code rate in joint source-channel coding

A practical method of adaptive rate allocation to source and channel codings for an independent loss channel is proposed for Internet video. It is based on the observations that the values of residual loss probabilities at the optimal code rates for different packet loss probabilities are closely clustered to the average residual loss probability for a transmission frame size n in RS(n,k) code and for a total bit rate R. These observations are then exploited to find the code rate for maximum PSNR. Simulation results demonstrate that the proposed method achieves a near-optimal bit-rate allocation in the joint source-channel coding of H.263 and RS(n,k) codings.

This paper proposes an adaptive rate control algorithm for MPEG-1 to MPEG-4 transcoder for video transmission over networks of different capacities and characteristics. It is based on an adaptive R-Q model to follow the new target bit-rate of a transcoded bitstream. In R-Q modeling, the R-Q characteristics predicted from a previous picture of the same picture type are used in combination with the R-Q data measured for the current picture. To overcome the quantizer mismatch between MPEG-1 and MPEG-4 syntax, a gradual adaptation function is applied. It is shown that the adaptive R-Q model-based rate control can efficiently control the output bit-rate to match the target bit budget.

This letter presents a fast motion vector refinement (MVR) for video downsampling in the discrete cosine transform (DCT) domain. To reduce the computational complexity of the MVR, we propose a fast search algorithm, based on minimizing the number of required checkpoints, and a computationally efficient method for extracting the motion-compensated DCT (MC-DCT) block. Compared to the known extraction method of the MC-DCT block, the proposed method requires fewer than one-half the multiplications and additions.

In this paper, we introduce an architecture of a video transcoder for converting MPEG-1 video to MPEG-4 simple profile with lower spatial resolution in the DCT-domain. For the DCT-domain transcoding to lower spatial resolution pictures, DCT-domain down-sampling is applied and a base motion vector (BMV) for the down-sampled MPEG-4 macroblock is to be calculated from the input motion vectors operating on the higher spatial resolution image. Quality can be significantly improved by refining the BMV. Starting with the BMV, the motion vector refinement (MVR) scheme searches for a refinement motion vector. We propose a fast MVR scheme for video down-sampling in the DCT-domain, based on minimizing the number of required check points and a computationally efficient method for extracting a motion compensated DCT (MC-DCT) block. We also show an efficient scheme for selecting a coding mode for the macroblock in the lower resolution video from those of the corresponding higher resolution video.

In this paper, we propose a new requantization method for the transcoder from MPEG-1 to MPEG-4 in DCT-domain. For rate control of heterogeneous transcoder, we use the requantization method. Conventional requantization methods for the homogeneous transcoder cannot be used directly for the heterogeneous transcoder due to the mismatch in quantization parameters between MPEG-1 and MPEG-4 syntax and the difference of compression efficiency between MPEG-1 and MPEG-4. In order to solve these problems, we propose a new requantization method for MPEG-1 to MPEG-4 transcoder. It consists of the conventional R-Q model with a simple feedback and an adjustment of quantization parameters with which the output rate of MPEG-4 follows the target rate of MPEG-1. By a simulation of transcoding, it is shown that the proposed requantization can generate output rate much closer to the target rate than the conventional requantization and it gives the superior image quality to the conventional requantization.

An adaptive video frame rate control method is proposed for the network of time-varying rate channel with explicit rate feedback. It consists of a prediction module of future channel rate and an adaptive frame discarding and skipping module. We derive an encoder buffer constraint which guarantees an end-to-end delay bound of video frames. The recursive least-squares (RLS) method is used as a tool to predict the low frequency component of channel rate. The adaptive frame discarding method prevents delay violation of frames due to the channel rate prediction error. Also, the frame skipping method adapts the encoder output rate to the channel rate while keeping a constant level of video quality. From the simulation result, it is shown that the proposed method keeps the decoded image quality nearly constant with much less number of skipped frames and no delay violating frame during the congestion period.

Recently, a number of empirical studies have demonstrated the existence of long-range dependence (LRD) or self-similarity in VBR video traffic. Since previous LRD models cannot capture all short- and long-term correlation and rate-distribution while still retaining mathematical tractability, there exist many doubts on the importance of SRD, LED, and rate-distribution on traffic engineering. In this paper, we present a video traffic model based on the shifting-level (SL) process with an accurate parameter matching algorithm for video traffic. The SL process captures all those key statistics of an empirical video trace. Also, we devised a queueing analysis method of SL/D/1/K, where the system size at every embedded point is quantized into a fixed set of values, thus the name quantization reduction method. This method is different from previous LRD queueing results in that it provides queueing results over all range not just an asymptotic solution. Further, this method provides not only the approximation but also the bounds of the approximation for the system states and thus guarantees the accuracy of the analysis. We found that for most available traces their ACF can be accurately modeled by a compound correlation (SLCC): an exponential function in short range and a hyperbolic function in long range. Comparing the queueing performances with C-DAR(1), the SLCC, and real video traces identify the effects of SRD and LRD in VBR video traffic on queueing performance.

Variable bit-rate (VBR) MPEG video traffic is highly bursty due to group of pictures structure, and shows time-variant statistical characteristics due to scene changes. These characteristics make it more difficult to manage network resources, and leads to the significant reduction in network utilization. We deal with the issues related to efficiently transmitting VBR MPEG video traffic over asynchronous transfer mode networks, while maintaining consistent visual quality and improving network utilization on the basis of real-time applications. First, we propose a joint encoder and channel-rate control scheme that comply with not only negotiated traffic parameters, but also constraints imposed by encoder and decoder buffers. Second, we propose a dynamic bandwidth renegotiation method by combining the above scheme with a traffic smoothing method that can make the peak rate close to the sustainable rate. The efficiencies of the proposed methods are compared with a few other competitive schemes such as transmission methods for unconstrained VBR and the constant bit-rate scheme with average bandwidth equal to that of the proposed methods.

Overlapped block motion compensation (OBMC) has been studied as a partial solution to the blocking artifact problems of the conventional non-overlapped block motion compensation (BMC). OBMC has been shown to provide improved prediction efficiency as well as better subjective quality for coded video compared with BMC. Also, by employing motion estimation techniques optimized for OBMC, called overlapped block motion estimation (OBME), we can obtain further substantial gains in compensation accuracy over conventional nonoverlapped block motion estimation (BME). However, OBME is too complex for practical realization because multiplication is required for each pixel of every candidate block in matching computations. In this work an effective motion estimation method is presented which significantly reduces the computational complexity of OBME. The proposed method uses a predetermined pattern in selecting matched pixels within a window support to reduce the multiplications of OBME. Simulation results show that the proposed method achieves about the same prediction performance as the conventional OBME.

In this paper, we propose a new method to reduce coding artifacts in transform image coding. It consists of a filtering scheme in spatial domain followed by a new projection algorithm onto the narrow quantization constraint set (NQCS). Generally, the projection onto NQCS can provide a better performance in terms of PSNR than the projection onto the conventional QCS. When the value of the narrowing factor is low, however, the projection onto NQCS does not guarantee a better performance, because it projects the transform coefficients that lie inside of the Original QCS onto NQCS. The proposed method projects transform coefficients that lie only outside of the original QCS onto the NQCS. Different projection methods are used for AC and DC coefficients, because the characteristics of distribution are different. The PSNR improvement ranges between 0.7 and 1.4 dB, compared with the decoded images without postprocessing. The PSNR gain of 0.2 to 0.5 dB is obtained, compared with the conventional projection method onto the NQCS.

The arrangement of the I-picture starting times of multiplexed variable bit rate (VBR) MPEG videos may significantly affect the cell loss ratio (CLR) characteristics of superposed traffic. In this paper, we deal with the problems due to the starting time arrangement of VBR MPEG videos. VBR MPEG video traffic is modeled by a sequence with time-varying and periodic picture-type dependent rate envelopes. From extensive investigations into the relationships between the starting time arrangement and the queueing performance, it is shown that the average power of superposed VBR MPEG video traffic can be a good measure for the burstiness of the traffic. Then, we can derive a starting time selection method for a newly requested VBR MPEG video that can minimize the CLR as well as the peak cell rate of the superposed traffic including the new request itself, and an efficient scheduling method called MC-scheduling is also proposed as an application of the starting time selection method. The exactness and efficiencies of the proposed methods are shown by comparing them with other scheduling methods in terms of the smoothness and the CLR performances.

The arrangement of I-picture starting times of multiplexed VBR MPEG videos may significantly affect the cell loss characteristics of superposed traffic. This paper presents a starting time selection method of a new request so that the minimum and maximum cell loss ratios (CLRs) of superposed traffic including the new request are expected. For this purpose, both single and superposed VBR MPEG video traffic are modeled as a time-varying periodic picture-type dependent rate envelope sequence. Through experiments and observations, we show that the average power of superposed VBR MPEG video traffic is a good measure for the burstiness of the traffic. The proposed starting time selection method is derived from these properties. The efficiencies of the proposed method are tested by comparison with other scheduling methods for the smoothness of the peak rate and the CLRs of the scheduled-superposed stream, and, we also provide connection admission control (CAC) methods applicable to ATM multiplexers which can manipulate the starting times of sources and video on demand (VOD) servers which can do that.

Some complicated features of VBR MPEG video traffic such as the different statistical properties according to different picture coding types and the periodic peaks due to I-pictures in the MPEG coding make it difficult to manage resources in ATM networks. In this paper, we discuss some problems of usage parameter control (UPC) and connection admission control (CAC) for VBR MPEG video sources in ATM networks, and propose an efficient traffic control framework for overcoming these problems. First, we model a VBR MPEG video traffic at slice level. Using this model, we analyze an ordinary leaky bucket (LB) system and present some shortcomings of the LB system for policing VBR MPEG video sources. To overcome these shortcomings, we propose a new UPC mechanism called the picture-type dependent LB (PDLB) system. We also discuss a key factor characterizing the cell loss characteristics of ATM multiplexer called 'the periodic-peak position effect'. To reflect this effect on CAC, we propose a new CAC method in cooperation with the PDLB mechanism.

In this paper, a new hierarchical edge-based block motion estimation scheme is proposed for the exact motion estimation in the passband structure. The basic idea is to estimate an initial motion at the top-most layer in order to increase the accuracy of the motion displacement at the next lower layers in the motion estimation pyramid. This is based on a blockmatching method by the block-edge pattern. In addition, we present a technique of classifying a block- edge pattern by use of only two parameters, which are obtained by calculating two highpass subband signals at the top-most layer in the motion estimation pyramid. Using this scheme, the coding performance is improved by the exact motion estimation. We also present a method to encode the residual error frames in a passband structure. Simulation results show that the performance of the scheme with the edge-based initial motion estimation is better than without it, especially when a high degree of motion exists.

In this paper, a new technique for designing linear phase QMF filters is proposed. These filters have the frequency responses of sharp and symmetrical transition charcteristics, and the capabilities of perfect reconstruction and simple implementation. By applying video signals into basic sub-band systems, the performances between designed QMF pair and other QMF pairs1,2 are compared in alias-cancellation!reconstruction errors and Peek Signal- to-Noise Ratios(PSNRs). It is shown that the proposed QMF pair has better performance than the short kernel filter pair1, and similar performance as the long kernel filter pair2.

The result of a spatio-temporal gradient (STG) method with generalized least square estimation (GLSE) is presented for the detection of an object motion in an image sequence corrupted by additive white noise. The proposed method is applied to an automatic target tracker using a high speed 16-bit microprocessor in order to track one moving target in real time. Experimental results show that the proposed method has much better performance over the conventional one with least square estimation (LSE).>

Techniques for varying duobinary and modified duobinary, which are two of the most important partial-response codes, to limit their run lengths without destroying their minimum bandwidth property, are introduced. One of the example codes is the first run-length-limited variant of duobinary ever known, and the other two are already publicized variants of modified duobinary, but are described either in more detail or from a different point of view. Performance figures of the three codes such as power spectrum, eye width, and error probability are experimentally confirmed to be acceptable for all engineering purposes.

By generalizing a concept on a power spectral zero at the Nyquist frequency, a condition for the line codes' capability to signal with the minimum bandwidth is established. From the condition, lower bounds of eyewidth are obtained for various values of a new system parameter called intersymbol interference sum variation (ISV). As an application of the condition, the nonminimum-bandwidth bipolar code is modified to a minimum-bandwidth code, which can be directly used in conventional bipolar signaling systems with increased bandwidth efficiency.

The conditional runlength code is known to be optimum for the Markov source model, but it is complicated in implementation and not efficient for practical line-based transmission. To avoid these disadvantages, a conditional incremental-runlength coding algorithm based on the same model is investigated. The entropy of the incremental runs is shown to be less than that of the runs, but additional information is required for the state identification. Yet, for a practical case with two composite states, the required state information is reduced so much that the new coding is shown to be better than the conditional runlength coding in line-based transmission. The coding efficiency is compared with some other well-known codes for the standard CCITT test documents. It is found to be one of the best codes, especially for the documents of high complexity.

In recent years adaptive linear estimation based upon the gradient-following algorithm has been proposed in a wide range of applications. However, little analysis on the convergence of the estimation has appeared when the elements of the data sequence are dependent. This paper presents such an analysis under the assumptions of stationarity andM-dependence (all data sets separated by more than a constantM < \inftyare statistically independent). It is shown that for a sufficiently small adaptation constant, the mean error in the estimator weights converges to a finite limit, generally nonzero. In addition, hounds on the norm of the mean weight-deviation and on the mean norm-square of the weight-deviation are found and shown to converge to asymptotic bounds, which can be made arbitrarily small by decreasing the adaptation constant and increasing the data block length over which gradient estimates are made.