Third generation full-waveform (FW) LIDAR systems collect time-resolved 1D signals generated by laser pulses reflecting off of intercepted objects. From these signals, scene depth profiles along each pulse path can be readily constructed. Using the conventional sampling process, however, massive amounts of data are typically required in order to achieve acceptable depth and spatial resolutions, and this data must be stored, transmitted, and processed. We have previously shown that such signals can be sampled at sub-Nyquist rates by using a finite rate of innovations (FRI) model. That work, however, used LIDAR data for which ground-truth distances were not available and it was therefore not possible to fully evaluate the range precision of an FRI-based representation. Here, we apply the proposed methodology to carefully ground-truthed laboratory data in order to better characterize its capabilities and limitations.

The human brain is a complex and dynamic system. This paper quantifies how it responds to change in video quality through changes in the distribution of feature vectors extracted from high-resolution electroencephalograph (EEG) signals. Specifically, subjects watch test video sequences with and without degradation while their brain response is recorded using a 128-channel EEG system. Power band feature vectors are then formed for each channel with each power band appearing to follow a log-normal distribution. Greater mean and variance values are observed over the occipital lobe in the 1-4Hz and 4-8Hz range as video quality decreases, indicating a change in neural firing as the video becomes harder to visually process.

In this work we examine the best system for subject identification and authentication using the electroencephalograph (EEG) signals that result when video stimuli are presented. A comparison is done on the performance of linear support vector machine (SVM) and Gaussian Mixutre Models (GMM) using cepstral coefficient feature vectors for subject classification. We evaluate and identify which region of the head results in the best classification and authentication results using a GMM ratio test between the claimed subject and a background model.

Third-generation full-waveform (FW) light detection and ranging (LIDAR) systems collect time-resolved 1-D signals generated by laser pulses reflected off of intercepted objects. From these signals, scene depth profiles along each pulse path can be readily constructed. By emitting a series of pulses toward a scene using a predefined scanning pattern and with the appropriate sampling and processing, an image-like depth map can be generated. Unfortunately, massive amounts of data are typically acquired to achieve acceptable depth and spatial resolutions. The sampling systems acquiring this data, however, seldom take into account the underlying low-dimensional structure generally present in FW signals and, consequently, they sample very inefficiently. Our main goal and focus here is to develop efficient sampling models and processes to collect individual time-resolved FW LIDAR signals. Specifically, we study sub-Nyquist sampling of the continuous-time LIDAR FW reflected pulses, considering two different sampling mechanisms: 1) modeling FW signals as short-duration pulses with multiple band-limited echoes; and 2) modeling them as signals with finite rates of innovation.

Third generation LIDAR full-waveform (FW) based systems collect 1D temporal profiles of laser pulses reflected by the intercepted objects to construct depth profiles along each pulse path. By emitting a series of pulses towards a scene using a predefined scanning pattern, a 3D image containing spatial-depth information can be constructed. Achieving super-resolution to resolve finer spatial details is of great interest because the spatial resolution of a LIDAR system is typically limited by the size of the spot on the target. In this study, we consider the problem of resolving range maps to resolutions smaller than the size of the spot using overlapping spots. This overlap provides the additional information needed to locate multiple objects within a spot using sparse source separation, thus achieving super-resolution.

In this paper, we address the problem of sampling the LIDAR range map. The significance of this problem is based upon the fact that large datasets generated by sampling inefficiently impose storage, processing and transmission limitations. Current compression approaches addressing these issues rely on collecting large amounts of data to analyze and throw away the redundancies. Unfortunately, the sampling performed by these approaches is still inefficient. Our approach to compression consists instead, on using a random uniform impulsive scan sampling scheme with sampling densities depending on the surface complexity. Turns out that the number of scanned time-resolved waveforms required to achieve “good” approximations, needs to satisfy the bound equation, where equation is an estimator of the surface numerical rank. Such a bound allows one to sub-sample surfaces which are close to a low-rank space and update, on the fly, the sampling densities required to improve the quality of the approximation.

In this paper, we consider the problem of characterizing the quality of audio that is impaired by time-varying distortions. Specifically, the quality of such audio must be characterized by a time-dependent function. We first study issues associated with the human subjective testing which must be addressed in order to create such a function, and we conclude that the best approach for achieving both high temporal resolution and high accuracy is most likely a combination of two testing methodologies. Using the collected subjective data, we then design an objective metric that attempts to replicate the subjective responses. The proposed metric uses as its core a subset of the model output variables that are part of the ITU perceptual evaluation of audio quality (PEAQ) recommendation as well as the structural similarity measure and the segmental signal to noise ratio. Depending on the dataset used in its design, we find that the estimated quality has an average correlation coefficient relative to the ground truth curves of between 0.887 and 0.933. Furthermore, we also find that it is possible to reduce the complexity of this metric considerably while only degrading its performance by between 0.3% and 1.6%. In addition, the reduced complexity metric appears to be more robust than the original one.

Polarization, a property of light that conveys information about the transverse electric field orientation, complements other attributes of electromagnetic radiation such as intensity and frequency. Using multiple passive polarimetric images, we develop an iterative, model-based approach to estimate the complex index of refraction and apply it to target classification.

One of the major problems associated with LIDAR sensing is that significant amounts of data must be collected to obtain detailed topographical information about a region. Current efforts to solve this problem have focused on designing compression algorithms which operate on the collected data. These, however, require the collection of large amounts of data only to discard most of it in some transformed domain. Instead, compressive sensing has demonstrated that highly accurate signal reconstructions are achievable even when sampling below the Nyquist rate. Such sensing is clearly desirable for LIDAR range data compression if it can be achieved. One notes, however, that compressive sensing requires a priori knowledge of the sparsifying basis of the signal which is a major problem for LIDAR since that basis depends not only on the underlying scene complexity but also on the laser spot size and target distance. For these reasons, the goal of this research is to take the first steps in establishing a relationship between typical LIDAR scenes of varying complexity and the sparsity of the scene compressively sampled.

Imaging polarimetry has emerged as a powerful tool for application in the field of remote sensing. In this paper, we present a novel technique for estimating the surface normal angle of each of the individual facets of a target object using passive polarimetric data. The passive polarimetric imaging system described here uses multiple measurements of the output Stokes vectors along with the reflection Mueller matrix, to extract the surface normal angle corresponding to individual facets of the target object. The knowledge of this parameter is indispensable for determining the orientation and surface geometry of the target object and thus facilitates applications like object recognition, shape extraction and building scene geometry. The worst-case error is found to be less than 2%, based on Monte Carlo computer simulation results.

In this paper, we develop a low complexity algorithm for spatial overlap detection and characterization that operates directly on the bitstream of motion-JPEG compressed video. Its low complexity and the fact that it does not require video decoding at the sensor nodes make it well suited to multi-view distributed video coding applications for wireless sensor networks.

Efficient transmission and storage of digital audio signals can be accomplished using a wide variety of compression algorithms. To compare and optimize the performance of such algorithms, objective metrics are often used to measure the quality of such compressed audio signals since subjective testing is extremely time consuming. In this paper, we consider the application of the structural similarity measure, originally developed for image quality assessment, to the problem of audio quality assessment. Specifically, we study two different implementations of the structural similarity index: the first applies it to short and fixed time-domain frames of an audio sequence while the second decomposes the audio signals into a non-redundant, time-frequency map and then compares the structural similarity in the resulting 2-dimensional domain. We compare the accuracies of the two structural similarity measures with those of other accepted objective audio quality metrics relative to MUSHRA-based subjective audio evaluations.

The goal of this paper is to develop an audio quality metric that can accurately quantify subjective quality over audio fidelities ranging from highly impaired to perceptually lossless. As one example of its utility, such a metric would allow scalable audio coding algorithms to be easily optimized over their entire operating ranges. We have found that the ITU-recommended objective quality metric, ITU-R BS.1387, does not accurately predict subjective audio quality over the wide range of fidelity levels of interest to us. In developing the desired universal metric, we use as a starting point the model output variables (MOVs) that make up BS.1387 as well as the energy equalization truncation threshold which has been found to be particularly useful for highly impaired audio. To combine these MOVs into a single quality measure that is both accurate and robust, we have developed a hybrid least-squares/minimax optimization procedure. Our test results show that the minimax-optimized metric is up to 36% lower in maximum absolute error compared to a similar metric designed using the conventional least-squares procedure.

A perceptually scalable audio coder generates a bit-stream that contains layers of audio fidelity and is encoded in such a way that adding one of these layers enhances the reconstructed audio by an amount that is just noticeable by the listener. Such algorithms have applications like music on demand at variable levels of fidelity, for instance using 3G and 4G cellular radio systems operating at different bit rates. While the MPEG-4 natural audio coder can create finely scalable bit streams using bit sliced arithmetic coding (BSAC), its perceptual quality at low bit rates is poor. On the other hand, the nonscalable transform-domain weighted interleaved vector quantization (TWIN-VQ) performs well at low bit rates. In this paper, we present a modified version of TWIN-VQ algorithm that generates a perceptually scalable bit-stream with many fine layers of audio fidelity. Using TWIN-VQ as our base ensures the best possible perceptual quality at low bit rates. Specifically, the proposed scalable algorithm performs as well as TWIN-VQ at rates of 8 to 16 kb/s and outperforms scalable BSAC by between 64% and 172% at rates of less than 24 kb/s.

Passive imaging polarimetry has emerged as an useful tool in many remote sensing applications including material classification, target detection and shape extraction. In this paper we present a method to classify specular objects based on their material composition from passive polarimetric imagery. The proposed algorithm is built on an iterative model-based method to recover the complex index of refraction of a specular target from multiple polarization measurements. The recovered parameters are then used to discriminate between objects by employing the nearest neighbor rule. The effectiveness of the proposed method is validated with data collected in laboratory conditions. Experimental results indicate that the classification approach is highly effective for distinguishing between various targets of interest. Most significantly, the proposed classification method is robust to a wide range of observational geometry.

A perceptually scalable audio coder generates a bit-stream that contains layers of audio fidelity and is encoded in such a way that adding one of these layers enhances the reconstructed audio by an amount that is just noticeable by the listener. Such algorithms have applications like music on demand at variable levels of fidelity for 3G and 4G cellular radio systems operating at different bit rates. While the MPEG-4 natural audio coder can create scalable bit streams, its perceptual quality at low bit rates is poor. On the other hand, the non scalable TWIN-VQ performs well at low bit rates. In this paper we present a technique to modify the TWIN-VQ algorithm such that it generates a perceptually scalable bit-stream with layers of audio fidelity. Using the TWIN-VQ as our base ensures the best possible perceptual quality at low bit rates (8 - 16 kbps).

ITU-R BS.1387-1 gives a method for objective measurement of perceived audio quality known as PEAQ (perceptual evaluation of audio quality). This algorithm has been developed for measuring the quality of mid and high quality audio. We show that the advanced version of PEAQ performs poorly when compared to the previously developed energy equalization approach (EEA) for evaluating the quality of low bitrate scalable audio. We also show that including an energy equalization parameter as one of the model output variables (MOVs) of the advanced version improves its performance significantly; the performance of this modified version is superior to that of EEA.

In this paper, we study coding artifacts in MPEG-compressed scalable audio. Specifically, we consider the MPEG advanced audio coder (AAC) using bit slice scalable arithmetic coding (BSAC) as implemented in the MPEG-4 reference software. First we perform human subjective testing using the comparison category rating (CCR) approach, quantitatively comparing the performance of scalable BSAC with the nonscaled TwinVQ and AAC algorithms. This testing indicates that scalable BSAC performs very poorly relative to TwinVQ at the lowest bitrate considered (16 kb/s) largely because of an annoying and seemingly random mid-range tonal signal that is superimposed onto the desired output. In order to better understand and quantify the distortion introduced into compressed audio at low bit rates, we apply two analysis techniques: Reng bifrequency probing and time-frequency decomposition. Using Reng probing, we conclude that aliasing is most likely not the cause of the annoying tonal signal; instead, time-frequency or spectrogram analysis indicates that its cause is most likely suboptimal bit allocation. Finally, we describe the energy equalization quality metric (EEQM) for predicting the relative perceptual performance of the different coding algorithms and compare its predictive ability with that of ITU Recommendation ITU-R BS.1387-1.

We consider here the implementation of efficient search and retrieval of digital elevation maps (DEMs), in particular the ability to conduct elevation searches without decompressing the entire map. We utilize set partitioning in hierarchical trees (SPIHT) for the compression, and compare two wavelet-based search and retrieval systems: one utilizing nonlinear max- and min-lifted integer wavelets and the other using the standard 5/3 integer wavelet. While the coarse-scale maxima/minima preservation inherent in the max- and min-lifted wavelets may seem ideal for the implementation of efficient elevation searches, the interscale propagation of maxima/minima and the bit-plane layering of the SPIHT coder adversely affect the efficiency of such searches. The use of the standard 5/3 integer wavelet requires the addition of separate maxima and minima information but does not have the complication of coefficient propagation, providing more satisfactory results. Elevation search bitrates are presented for both systems and compared to results obtained using the Kakadu implementation of the JPEG2000 standard.

In this paper we consider the use of nonlinear max- and min-lifted integer wavelets encoded using set partitioning in hierarchical trees (SPIHT) for the compression and efficient data-retrieval of digital elevation maps (DEMs). Max- and min-lifted integer wavelets losslessly code data such that the regional maxima and minima are preserved at the coarsest scale, which facilitates efficient multiresolutional elevation searches. Of particular interest in this application are the effects of maxima/minima propagation between resolution scales, as well as the effects of the bit-plane layering of wavelet coefficients inherent in the SPIHT coder. This implementation of a DEM browsing system will be compared to one utilizing a standard 5/3 wavelet, such as is used in JPEG2000.

We study coding artifacts in MPEG-compressed scalable audio. Specifically, we consider the MPEG advanced audio coder (AAC) using bit slice scalable arithmetic coding (BSAC) as implemented in the MPEG 4 reference software. First, we perform human subjective testing using the comparison category rating (CCR) approach, quantitatively comparing the performance of scalable BSAC with the nonscalable TwinVQ and AAC algorithms. This testing indicates that scalable BSAC performs very poorly relative to TwinVQ at the lowest bitrate considered (16 kb/s), largely because of an annoying and seemingly random mid-range tonal signal that is superimposed onto the desired output. In order to understand better and quantify perceptually the various forms of distortion introduced into compressed audio at low bit rates, we apply two analysis techniques: Reng probing and time-frequency decomposition. The Reng probing technique is capable of separating the linear time-invariant component of a multirate system from its nonlinear and periodically time-varying components. Using this technique, we conclude that aliasing is probably not the cause of the annoying tonal signal; instead, time-frequency analysis indicates that its cause is most likely suboptimal bit allocation.

In this work, we combine the recently proposed GLICBAWLS (grey level image compression by adaptive weighted least squares) lossless image compression algorithm with integer-to-integer wavelet transforms to create a resolution-progressive algorithm capable of lossless quality. We find that modifying the prediction window to exploit inter-scale redundancy is quite helpful at some points in the processing but not universally. To evaluate the effectiveness of our algorithm, we consider the compression of both 16-bit digital elevation maps (DEMs) and 8-bit images. For purposes of comparison, we also apply a lossless version of SPIHT to the same data sets. For higher resolution DEMs, we find that GLICBAWLS+wavelets is superior while for lower resolution DEMs, SPIHT is the better choice.

We study the application of nonlinear max- and min-lifted wavelets to the compression of digital elevation maps (DEMs). Such wavelets have the desirable property that regional maximum or minimum values are preserved at coarse resolutions, making it possible with proper encoding to efficiently search for high/low points without decoding the entire file. In course of studying these nonlinear transforms, we first quantify the dynamic range expansion incurred by using them, and we then illustrate with an example how maximum values propagate to coarse scales. As a first step, we consider encoding the resulting coefficients using two popular algorithms: set partitioning in hierarchical trees (SPIHT) and stack-run. In doing this, we find that efficiently exploiting inter-scale redundancies in the transform is far more important for these wavelets than it is for the more commonly used (2,2)-integer wavelet (also called the 5/3 wavelet).

An embedded coding algorithm creates a compressed bit stream that can be truncated to produce reduced resolution versions of the original image. Unfortunately, such algorithms are relatively slow. We introduce two new concepts in an effort to address this problem: adaptive embedding and cache-based zerotree processing.

Summary form only given. We consider here a new approach to successive coefficient refinement which speeds up embedded image compression and decompression. Rather than sending the binary refinement symbol typical of existing embedded coders, our algorithm uses a ternary refinement symbol, allowing the encoder to tell the decoder when its current approximation of a given wavelet coefficient is exact. Thus, both encoder and decoder operate faster because they process fewer refinement symbols, yet the fundamental structure of the refinement process remains unchanged, i.e. it still represents a binary subdivision of the uncertainty interval. To implement a complete encoder, we combine the proposed refinement process with Shapiro's embedded zerotree wavelet (EZW) algorithm. Results for lossless compression are shown. Without optimization, the speed increase is between 5 and 12%; with optimization, it is between 9 and 15%.

We consider here a new approach to successive coefficient refinement which speeds up embedded image compression and decompression. Such speedups are especially useful when lossless or near lossless image compression is required because so many refinement bits must be transmitted. Rather than sending a binary refinement symbol typical of existing embedded coders, our algorithm uses a ternary refinement symbol, allowing the encoder to tell the decoder when its current approximation of a given wavelet coefficient is exact. Thus, both encoder and decoder operate faster because they process fewer refinement symbols, yet the fundamental structure of the refinement process remains unchanged-something that is not true if a non-binary refinement tree is used instead. To implement a complete encoder, we combine the proposed refinement process with Shapiro's embedded zerotree wavelet (EZW) algorithm. By optimizing this process, speed increases of between 13 and 16% can be realized on the set of test images.

To reduce the complexity of a video encoder, we introduce a new approach to hybrid DPCM-transform video compression in which pan compensation is performed outside the feedback loop. While the basic idea is conceptually similar to the pan compensation algorithm proposed by Taubman and Zakhor (1994) for their 3D subband coder, our method is different in that it continually tracks and updates the image in the feedback loop in the same way as a conventional hybrid coder. Using both residual energy and reconstruction error as metrics, we show that pan compensation implemented outside the feedback loop compares very favorably to similar compensation implemented within the conventional hybrid-transform framework. Furthermore, if the spatial coder used to compress the residual images outputs an embedded bit stream, then the complete system is spatially scaleable.

To reduce the complexity of a video encoder, we introduce a new approach to global motion compensation related to the conventional hybrid DPCM-transform method in which the motion compensation is performed outside the feedback loop. This method is potentially useful in a variety of situations, but we focus here on an area of great interest to us- its application to the problem of video compression from remote airborne platforms. While the basic idea is conceptually similar to the pan compensation proposed by Taubman and Zakhor for their 3D subband coder, our method continually tracks and updates the image in the feedback loop in the same way as a conventional hybrid system. Using both residual energy and reconstruction error as metrics, we show that the new motion compensation scheme compares very favorably to the conventional one. We also discuss how one can achieve subpixel interpolation within the proposed framework.

We propose a wavelet-based image compression algorithm that achieves robustness to transmission errors by partitioning the transform coefficients into groups and independently processing each group using an embedded coder. Thus, a bit error in one group does not affect the others, allowing more uncorrupted information to reach the decoder.

We present a new family of image compression algorithms derived from Shapiro's embedded zerotree wavelet (EZW) coder. These new algorithms introduce robustness to transmission errors into the bit stream while still preserving its embedded structure. This is done by partitioning the wavelet coefficients into groups, coding each group independently, and interleaving the bit streams for transmission-thus, if one bit is corrupted, then only one of these bit streams will be truncated in the decoder. If each group of wavelet coefficients uniformly spans the entire image, then the objective and subjective qualities of the reconstructed image can be very good even when an error occurs. To illustrate the advantages of this new family, we compare the performance of some of its members to that of the conventional EZW coder. For example, one of the robust variations has a peak signal to noise ratio (PSNR) slightly lower than the conventional algorithm when no errors occur (-0.12 dB for Lena and -0.5 dB for Barbara, both coded at 1.0 bits/pixel), but when a single error occurs at bit 1000, the PSNR of the new coder is more than 5 dB higher for both test images. The PSNR versus the bit error position is shown using two different partitionings of the wavelet coefficients and four interleaved bit streams.

We consider the problem of implementing perfect reconstruction modulated polyphase filter banks having infinite impulse response (IIR) subfilters in systems that must process infinite length 1-D inputs. This task is complicated by the fact that the synthesis bank needed to ensure perfect reconstruction in such a system generally has noncausal or reverse-time filter sections. To overcome this limitation, we develop a number of unique buffering schemes that make possible the implementation of such reverse-time filters with infinite length signals. For the important special case when the polyphase subfilters are allpass, we also present a method that allows for very efficient implementation of tree-structured filter banks and wavelet transforms. We then describe a wideband coding algorithm based largely on work done by Lokhoff (1991) and Johnston (1980, 1988). Using this algorithm, we compare the performance of IIR filter banks with that of more conventional pseudo-QMF and tree-structured QMF banks. The results of these comparisons indicate that one of our filter banks-a tree-structured allpass bank-achieves slightly better performance than equivalent FIR banks in terms of rate and distortion (both objectively and subjectively measured) while requiring lower computational complexity.

This paper re-examines modulated polyphase filter banks which use reverse-time subfilters to achieve perfect reconstruction and studies their performance for image coding when used to form a wavelet decomposition. To code the wavelet coefficients, we use the embedded zerotree wavelet (EZW) algorithm developed by Shapiro (1993) which is reasonably simple and yet achieves very good rate-distortion performance. The problem of eliminating filter transients is thoroughly studied and optimal results are derived in both the absence and presence of subband coefficient quantization. Using this analysis as a starting point, we develop a new method of coding the filter states (required to eliminate edge transients) for transmission to the receiver which efficiently exploits interscale redundancy and is easily incorporated into the zerotree algorithm. Comparing this direct transmission method to circular convolution, we find that it achieves superior rate-distortion performance in a wide range of cases. Coding comparisons between two infinite impulse response (IIR) wavelets and a number of biorthogonal wavelets are presented. These comparisons indicate that the performance of the polyphase allpass wavelet is comparable to that of the best finite impulse response (FIR) biorthogonal wavelets with considerably reduced computational complexity.

Compares for image coding applications a low-complexity IIR wavelet based on an all-pass polyphase decomposition to a pair of linear phase biorthogonal wavelets. To code the wavelet coefficients, the authors use Shapiro's (1993) zerotree algorithm which has the virtues of being both efficient and delivering excellent performance (in a rate-distortion sense). They consider a variety of methods for eliminating filter transients at the image boundaries including circular convolution, symmetric extension (for the biorthogonal wavelets), and direct transmission (for the IIR wavelet). By also coding the filter states in a zerotree form, they find that direct transmission generally performs better than circular convolution. Finally, they show that the use of this IIR wavelet provides equivalent performance to the biorthogonal wavelets at greatly reduced computational complexity.

Proposes a modification to the classic time-frequency permutation method of signal scrambling which reduces its complexity and memory requirements without decreasing its performance. This new approach is applicable to any filter bank or transform and requires only that the decomposition in question be implemented in a polyphase form. To minimize the computational complexity, modulated polyphase filter banks like the cosine-modulated and allpass filter banks can be selected. Using a two dimensional extension to the basic algorithm, the authors consider the problem of image scrambling and include simulation results.>