This paper considers a comprehensive and collaborative project to collect large amounts of driving data on the road for use in a wide range of areas of vehicle-related research centered on driving behavior. Unlike previous data collection efforts, the corpora collected here contain both human and vehicle sensor data, together with rich and continuous transcriptions. While most efforts on in-vehicle research are generally focused within individual countries, this effort links a collaborative team from three diverse regions (i.e., Asia, American, and Europe). Details relating to the data collection paradigm, such as sensors, driver information, routes, and transcription protocols, are discussed, and a preliminary analysis of the data across the three data collection sites from the U.S. (Dallas), Japan (Nagoya), and Turkey (Istanbul) is provided. The usability of the corpora has been experimentally verified with a Cohen's kappa coefficient of 0.74 for transcription reliability, as well as being successfully exploited for several in-vehicle applications. Most importantly, the corpora are publicly available for research use and represent one of the first multination efforts to share resources and understand driver characteristics. Future work on distributing the corpora to the wider research community is also discussed.

We investigate the uniqueness of driver behavior in vehicles and the possibility of using it for personal identification with the objectives of achieving safer driving, of assisting the driver in case of emergencies, and of being a part of a multi-mode biometric signature for driver identification. We use Gaussian mixture models (GMM) for modeling the individualities of the accelerator and brake pedal pressures, and focus on not only the static features, but also the dynamics of the pedal pressures. Experimental results show that the dynamic features significantly improve the performance of driver identification.

The delivery of the proper dose of radiation to a tumor volume without causing irreparable damage to healthy tissue and critical organs is paramount in radiation therapy. With the development of the multi-leaf collimator (MLC) a new type of conformal radiation therapy, known as intensity modulated radiation therapy (IMRT), was developed and deployed in a number of clinics around the world. The inverse optimization parallelism to the solution is the approach taken in IMRT and in this paper, where the optimality is based on the minimization of a cost function depending on the difference between the prescribed and the delivered dose. Until recently, Bayesian techniques have been used to solve this problem. Here we investigate and implement two new cost functions frequently used in signal processing community as part of the dose optimization. A simulation model was developed and tested using both phantom data and CT scan results from two patients with encouraging results.

This paper focuses on a class of speech enhancement systems using the amplitude spectral estimates of noisy speech and noise to drive a Wiener filter to suppress simultaneously the ensemble of degradations picked up by microphones. A simple stereo microphone set, with left and right channels can be used to provide enough separation that unwanted signals can be reduced significantly to yield an acceptable quality speech signal for hands-free telephony applications in a vehicular environment. A generalized transform approach was introduced and experimental results show great potential when using the DCT as alternatives to the traditional Fourier transform approach to derive the amplitude spectral estimates of the corrupted speech signals and the noise process.

This paper introduces a distributed object based collaboration system called Collaboard, which can be effectively used to conduct signal processing classes in an interactive fashion. Collaboard allows a group of users in a heterogeneous network environment to share multimedia objects, such as text, geometric entities, equations, images, audio/video objects, and OLE/sup TM/ objects. Collaboard supports multiple user groups and allows a user to actively participate in multiple learning groups. Matlab/sup TM/ has been integrated into this Collaboard as the computational workhorse. The Matlab programs or tasks initiated by any participant are replicated over the network to every participant. Our Collaboard architecture is a distributed object-based tool supporting object video and object audio. The current version of the system includes a RealAudio/sup TM/ server to support streaming audio capability. We believe that the comprehensive and user-friendly architecture of this Collaboard will be a very powerful working tool for DSP/communication systems classes in active learning environment.

This study proposes a novel autoassociator for associative storage of gray scale images. An autoassociator is a system capable of producing a complete pattern, when it is presented part of a learned pattern. Patterns in the proposed associative architecture are gray scale images with integer values. The system proposed here is superior to existing techniques using binary neural networks for associative storage of gray scale images in both the computational complexity and the processing unit requirements. The proposed system is parallel in nature and can be implemented using a synchronous or asynchronous parallel system with no added complexity.

In this study we propose a new architecture for texture classification based on pair-wise pixel associations as an extension of the recently developed multivalued recursive network (MAREN) architecture. Maybe more critically we propose a novel similarity measure and classification algorithm to be used with this network. The proposed fidelity criterion has been observed to be tightly coupled with the ubiquitous mean-square error (MSE) distance measure. Both system of associative relationships (SOAR) and MAREN structures can be considered extensions of the associative memory concept frequently used in neural networks. Our proposed similarity measure is based on the principle of directional divergence of interpixel relationships in a given texture and promises a number of advantages over the MSE measure. In this paper, SOAR will be discussed within the framework of a texture classification problem, but we believe it would be very easy to extend to other applications where interpixel relationship is the primary focus.

In this study, we present a new classroom environment to conduct digital signal processing and communication systems courses. Key features of the model are the collaborating instructor embracing students, a smart classroom equipped with a "whiteboard" and advanced telecommunication networks, electronic textbook, and other resources, World Wide Web (WWW), Matlab, and other online tools. The underlying assumptions of the educational process are team building instead of independent learning, collaborating/supervising instructor, lateral curriculum instead of a vertical curriculum, and idea-to-product design concept. We present a sample lecture in the proposed interactive classroom, where the concept of eye diagrams in regenerative repeaters are presented from the first author's text using Matlab and the WWW.

We present an algorithm which combines vector quantization (VQ) techniques with variable size block matching algorithms (BMA) for motion compensated image sequence coding. First, an algorithm which decomposes the image frame into blocks of different sizes with similar motion parameters is proposed. Since the algorithm is based on splitting the block that would result in the highest decrease in distortion, it does not require a merging step. Next, VQ techniques are utilized to expedite the search for the block motion vector without significant sacrifice in the temporal entropy reduction performance. We compare the performance and the computational requirements of the proposed algorithm with other BMAs.

Presents findings on image processing applications to the continuous and automatic monitoring and verification of the status of control devices and vehicle speed estimation on the Interstate-15 Reversible High Occupancy Vehicle (HOV) lanes as examples of "IVHS at Work". The overall goal of the study has been to supply additional enhanced monitoring capabilities for the HOV operations. These capabilities have been intended to assist, rather than to eliminate the human operators from the loop. The authors describe this unique undertaking together with the issues related to the systems architecture, hardware and software components, the integration, image processing tools, and preliminary field test results.

Code excited linear prediction (CELP) coding and its derivatives are currently the most frequently used techniques for speech compression at medium-to-low rate range. Until this study, the excitation vectors were always selected from a codebook generated by a Gaussian source or by the ensemble of residual signals collected from the used speech database. To the best of the author's knowledge, there is no study in which the excitation vectors were formed from a mixture of sources, a notion very successfully used by the speech recognition community within the hidden Markov model (HMM) framework. The authors proposed an improvement to the excitation model of CELP coders by embedding a labeled-state finite state vector quantization (FSVQ) and a mixture density approach in constructing 5-ms-long excitation vectors.>

The authors have investigated the stability problems observed in multichannel multidimensional linear predictive modeling of images. It is known that based on a positive definite autocorrelation matrix, singular values of the matrix H=delta(i+1)xHerm (delta(i+1)) must lie inside the unit circle for a stable solution, where delta(i+1 ) is the normalized partial correlation matrix and Herm denotes the Hermitian operator. The authors have developed a two-step stabilization method to obtain stabilized linear prediction coefficients for short term analysis windows formed digitized images. The authors have modified the multichannel Levinson recursion algorithm to include this stability procedure. They have tested the algorithm on numerous images commonly used in image coding and the results are very impressive.

In this study we have attempted to investigate the stability problems observed in multichannel multidimensional linear predictive modeling of images. Morf et al.[3] have shown that based on a positive definite autocorrelation matrix, singular values of the matrix H ? ?q + 1.HERM(?q + 1) must lie inside the unit circle for a stable solution, where ?q + 1 is the normalized partial correlation matrix and HERM(.) denotes the Hermitian operator. We have employed this stability method to modify the multichannel Levinson algorithm [1,2] for obtaining stable linear prediction coefficients. Since the procedure involved block-by-block processing of image intensity values, blocks of 32x32 pixels were defined as analysis windows. A two-step stabilization method has been developed for these windows and it is applied to the multichannel multidimensional linear prediction of monochromatic imagery. The first step is based on heuristic notions and employed for obtaining strictly positive definite multichannel autocorrelation matrices R[q]. The second step is based on forcing singular values of H to reside inside the unit circle for satisfying the stability criterion reported in [3].

We present a number of architectures for vector quantization (VQ) of speech and images using VLSI and VHSIC technologies. A Dual Distortion Processor Module (DPM) has been designed to compute the error vectors at a rate of 10 million vector operations per second in a systolic configuration. An array processor controller (APC) administers the system and determines the nearest neighbor matching codeword in either a full-search or a tree-search manner. A real-time system was built and tested in 0.5 bit per pixel (bpp) image coding application. We also present an architecture using VHSIC technology based on fuzzy associative memory (FAM) chips. In this case, the system has been configured in a VME bus environment and the overall number crunching task is handled by VHSIC chips.

This paper describes a composite source model for images. An image is segmented into uniform and homogeneous regions using centroid linkage region growing algorithms. The region homogenity is determined by the Student T-statistics. Excessive regions resulting from region growing are merged according to region merging rules. The initially segmented image is then clustered into classes with the help of the K-means algorithm. The image classes are shown in this paper as visual aids in judging the classification procedure. The class numbers and the corresponding pixel counts are also included. Finally, as an application of composite source models, an image is encoded using matrix quantizers with separate codebook for each class of the image.

Vector quantization (VQ), subband coding (SBC) and linear predictive coding (LPC) are three of the most effective data compression schemes used for medium-to-narrow band speech coding. In this study, we have attempted to improve the quality of encoded speech by using various combinations of these three coding methods. Waveform coders with rates 2400-9600 bits per second resulted in overall signal-to-distortion ratios of 6-12 dB. We have obtained somewhat lower values for segmented SNR's in the case of straight waveform encoding and higher values for the residually excited subband coded VQ quantizers as expected. However, informal listening tests yielded noticeable improvements over those of straight waveform VQ results. The best quality and the lowest transmission rates were achieved by a residually excited subband coded vector quantization system coupled with a three-way classifier with design parameters: LPC order P=14; 32-band complete binary tree QMF filter bank implementation of SBC and VQ waveform encoding with dimension K=32 at an overall bitrate of 3,100 bps.

This paper discusses a multi-stage matrix quantizer with a tree structure for image encoding wherein the number of searches required is a linear function of the product of the matrix size and the bit rate rather than an exponential one as in the case of a full search encoder. A two-step nine-way classification procedure is employed in order to preserve the edge contents of images. Further data rate reduction is achieved by preprocessing the images with a simple linear Hadamard transformer. A kl subblock of the input image is classified nine ways to detect the edge orientation. If it is not an edge a shade codebook is designed. Eight different codebooks are generated to accommodate each of the edge orientations under consideration. It is observed that the staircase degradation due to quantization is greatly reduced when edge detection is used. It is also found that the degradation due to tree encoding is only about 0.5 dB as compared to full search encoding.

This paper discusses the design of a matrix quantizer incorporating the human visual model for image encoding. It is well known that in image processing square error distortion measure is not the most suitable yardstick for the evaluation of subjective quality of reconstructed images. Since in almost all image data compression systems the human observer is the ultimate destination, one may take into account some parameters of the human visual system in computing the distance measure used in the design of an encoding algorithm. The matrix quantizer design and encoding algorithms have been modified to include a reasonably accurate model to reflect the human visual perception process. It is found that the reconstructed images in this case are sharper and exhibit much less staircase effect usually found in matrix quantizers that do not include such a model. The results were compared for rate one and 0.56 bits per pixel (bpp) with those of straight matrix quantizers of similar rate and size.

Vector Quantization (VQ) and subband coding (SBC) are two of the most efficient data compression systems in the field of medium-to-low rate speech waveform coding. In this paper, we examine the performance of a coding system operating at either 6,500 bits per second or 13,000 bits per second which incorporates a full-search vector quantizer to encode the outputs coming from a complete subband coder filter bank.

This paper discusses a matrix quantizer design algorithm for image encoding problems. The design algorithm is aimed at producing a codebook of matrices which are, at least, locally optimum with respect to a distortion measure. We have considered the squared error distortion measure in this work and generated codebooks based on a training sequence consisting of a number of pictures of different bit rates. The preliminary results show promise for further work in this direction.

A real-time full search vector quantization system for speech waveform coding is implemented using LSTTL and CMOS devices. The system consists of low-pass filters, A/D and D/A converters, an algorithm for discriminating voiced and unvoiced speed, a full search vector quantizer encoder and decoder, and a microprocessor-based controller. The system is designed to operate at two possible rates: one bit/sample using a dimension 8 vector quantizer (6500 bits/s) or 2 bits/sample using a dimension 4 vector quantizer (13 000 bits/s). In both cases the codebooks have rate 8 bits/vector. Separate codebooks were designed for voiced and unvoiced speech based on a training sequence of 640 000 samples containing five different speakers. The subjective and quantitative results are compared to both simulations and with a real-time array processor based implementation.

Vector quantizers of one and two bits per sample are designed for a training sequence of 640000 speech samples and tested on a speaker not in the training sequence. Both full search vector quantizers and tree search vector quantizers are considered. The tree searched codes are suboptimal in an information theory sense, but they have a greatly reduced search effort and provide a vector successive approximation quantizer.

An algorithm for the design of locally optimum vector quantizers relative to a distortion measure is used to design and simulate vector quantizers for both real sampled speech and for speech-like waveforms produced by a tenth order autoregressive random process with matching autocorrelation. Both squared-error and a weighted squared error were considered. The experimental results were compared with performance bounds from rate distortion theory based on the autoregressive model.

The design and evaluation of real-time implementable coding schemes for Gaussian multisource-multichannel communication systems are considered. The problem is to transmit the output from ann-dimensional Gaussian source over a memorylessm-dimensional Gaussian channel with power constraint and a mean-squared error distortion measure. Shannon performance bounds are obtained for this system through the optimum performance theoretically attainable (OPTA) relationship between the source distortion level\betaand the channel input power level\alpha. The optimum encoder is determined within the class of linear memoryless transformations, in which case the optimum decoder transformation becomes a least-mean-square estimator. The optimum linear encoder-decoder pair either coincides with or is very close to the Shannon bound in a number of significant cases.

Upper and lower bounds on first-order rate-distortion functions are derived for some selected speech probability models and two different fidelity criteria. The bounds are computed numerically when the analytical procedures are intractable. The models employed involve the Laplacian density, the gamma density, and first- and second-order modified Bessel functionsK_{o}andK_{1}. For both the absolute difference and the mean-square difference distortion measures, theK_{o}model yields the tightest bounds onR_{1}(D)functions.

The public telecommunication network of Turkey is briefly discussed in this paper. The existing telephone, telegraph, data, and other related services are summarized, and future trends in some of these services are indicated. These trends suggest that capital investments on the telecommunication services must be at least doubled in order that the standards in the country compete with those of the world averages. Finally, some of the research in communications is cited.