Vision-Language models (VLMs) integrate natural language understanding with visual data interpretation, crucial in diverse applications such as medical imaging. However, training VLMs on limited data, especially in radiology, remains a challenge. We propose a strategy to improve dual encoder performance under data constraints. Using contrastive learning to align visual and textual embeddings effectively, we generated a bag of rich textual descriptions using GPT-4 to augment merged information from esteemed medical resources and pre-trained BiomedCLIP. These rich textual descriptions provide in-depth information on disease visual description, major causes, and major symptoms, enhancing the model’s contextual understanding and classification accuracy. Unlike previous methods relying on a single alignment, our multiple alignment strategy associates multiple images with multiple textual descriptions per disease class while capping descriptors to maintain computational efficiency. Adapting the vision encoder for chest X-ray classification, our approach achieves competitive accuracy with fewer training pairs, highlighting its potential for data-limited domains.

Background: the process of cardiac segmentation using cardiac MRI images has been widely studied, and various deep learning models have been employed to address the complexities of heart chamber segmentation. Among these models, the 2-Dimensional (2D) UNet has demonstrated good performance in segmenting the left and right ventricles but has not been utilized to differential between the myocardium and papillary muscles. Consequently, researchers have proposed the use of the 3-Dimensional (3D) UNet as an alternative to the 2D UNet to improve segmentation outcomes. This study aims to compare the accuracy of 2D and 3D UNet models in segmenting the left ventricle using MRI images. Method: Both models were trained and tested on public ACDC dataset including 150 patients. Both models were trained for 140 epochs. To compare the accuracy of 2D and 3D UNet models, Dice Score Coefficient (DSC) and Hausdorff Distance (HD) were computed. Results: The 2D model achieved a mean Dice of 0.851 and a mean HD of 4.31 mm while the 3D UNet model achieved a higher performance in comparison with the 2D model with a mean Dice of 0.950 and a mean HD of 3.14 mm. Conclusion: The outcome of this study showed that 3D UNet is more suitable for the cardiac MRI segmentation.

Baggage screening in airports is a cornerstone in airport security measures. The advent of computer vision technologies in recent years has led to the development of several automated systems for identifying security threats in baggage scans. However, existing methods struggle to adapt to new threat categories when faced with a scarcity of data samples, and the rapid emergence of new threats. Hence, in this paper, we propose a novel CLIP-driven few-shot framework (CLIFS) to explore the potential of multi-modality using text-image fusion through contrastive learning to learn relevant contextual features for recognizing security threats with limited samples. By integrating features from GPT-4 generated captions with image features, CLIFS leverages both visual and textual data to significantly improve threat classification performance with limited samples in a few-shot learning context. Our proposed CLIFS was rigorously tested on the SIXray public available baggage X-ray dataset, where it outperformed state-of-the-art by 31.3% in accuracy and 28.40% in F1-score for the challenging 5-shots scenario, demonstrating its robustness and effectiveness in classifying threats from limited data samples.

Explainable AI (XAI) has revolutionized the field of deep learning by empowering users to have more trust in neural network models. The field of XAI allows users to probe the inner workings of these algorithms to elucidate their decision-making processes. The rise in popularity of XAI has led to the advent of different strategies to produce explanations, all of which only occasionally agree. Thus several objective evaluation metrics have been devised to decide which of these modules give the best explanation for specific scenarios. The goal of the paper is twofold: (i) we employ the notions of necessity and sufficiency from causal literature to come up with a novel explanatory technique called SHifted Adversaries using Pixel Elimination(SHAPE) which satisfies all the theoretical and mathematical criteria of being a valid explanation, (ii) we show that SHAPE is, infact, an adversarial explanation that fools causal metrics that are employed to measure the robustness and reliability of popular importance based visual XAI methods. Our analysis shows that SHAPE outperforms popular explanatory techniques like GradCAM and GradCAM++ in these tests and is comparable to RISE, raising questions about the sanity of these metrics and the need for human involvement for an overall better evaluation.

Cameras are now available at extremely low prices due to ongoing advancements in image acquisition hardware. However, the quality of images can be compromised by various distortions that occur throughout the entire process, from acquisition to processing and delivery. Over the past few decades, researchers have primarily focused on developing algorithms to assess the quality of distorted images. Unfortunately, certain distortions can also result from enhancement processes, such as over-enhancement and color saturation. Although there are metrics available for measuring contrast levels in images, there is currently no standard metric for evaluating the extent and effects of contrast enhancement. In this paper, we propose a new framework that expands the evaluation of contrast levels to ranking contrast-enhanced images. Our technique involves extracting a new set of features that accurately describe the effects of contrast enhancement. Furthermore, we integrate additional statistical indicators, such as skewness and kurtosis, which describe the degree of visual satisfaction linked to human perception. These identified characteristics are subsequently use with a simple classification module to determine the rank order for a given collection of contrast enhanced images. The results show excellent accuracy in correct ranking which outperforms state-of-the-art by more than $15 \%$.

This work develops a model-based compression scheme for seismic data. First, seismic traces are modeled as multitone sinusoidal waves superposition. Each sinusoidal wave is regarded as a model component and is represented by a set of distinct parameters. Second, a parameter estimation algorithm for this model is proposed accordingly. In this algorithm, the parameters are estimated for each component sequentially. A suitable number of model components is determined by the level of the residuals energy. Next, the residuals are compressed using entropy coding or quantization coding techniques. The corresponding compression ratios are presented. Finally, the proposed model-based compression scheme is compared with the linear predictive coding (LPC) algorithm and the distributed principal component analysis (DPCA) algorithm on a real seismic database. The performance of the proposed model based is shown to be superior to that of the LPC and DPCA.

Seismic data interpretation is a fundamental process in the pipeline of identifying hydrocarbon structural traps such as salt domes and faults. This process is highly demanding and challenging in terms of expert-knowledge, time, and efforts. The interpretation process becomes even more challenging when it comes to identifying multiple seismic events taking place simultaneously. In recent years, the technology trend has been directed towards the automation of seismic interpretation using advanced computational techniques and in particular deep learning (DL) networks. In this paper, we present our DL solution for concurrent salt domes and faults identification with very promising preliminary results obtained through applications to real world seismic data. The proposed workflow leads to excellent detection results even with small size training datasets. Furthermore, the resulting probability maps can be extended to even a larger number of structure types. Precisions of the order of more than 96% were obtained with real data when three types of seismic structures are present concurrently.

Annotating seismic data is expensive, laborious and subjective due to the number of years required for seismic interpreters to attain proficiency in interpretation. In this paper, we develop a framework to automate annotating pixels of a seismic image to delineate geological structural elements given image-level labels assigned to each image. Our framework factorizes the latent space of a deep encoder-decoder network by projecting the latent space to learned sub-spaces. Using constraints in the pixel space, the seismic image is further factorized to reveal confidence values on pixels associated with the geological element of interest. Details of the annotated image are provided for analysis and qualitative comparison is made with similar frameworks.

A plethora of Contrast Enhancement (CE) methods has been proposed in the literature. Each of these has its own strengths and limitations. Further, the quality of the resulting enhanced images depends upon the original image and its content. Hence, a given CE method can provide good quality for a certain image but a poorer quality for another. In this paper, we propose a novel workflow to provide an automatic ranking of enhanced images which may have been obtained using different techniques. The proposed technique is based on a Convolutional Neural Network (CNN) using saliency information. The idea is to start by comparing two enhanced versions of a given image in order to select the best one automatically based on perceived quality. Here, a saliency map is used to select relevant patches which are highly correlated with the human visual system sensitivity. The well-known Structural Similarity Image Metric (SSIM) map is also employed to compare the similarity between both enhanced images. Using such information, a CNN model is trained to predict the rank in terms of the image quality as perceived by humans. The algorithm is tested over three CE benchmarking databases with the experimental results validating the superiority of the proposed system as compared to state-of-the-art CE evaluation techniques.

In this paper, we present an approach for detecting faults within seismic volumes using a saliency detection framework that employs a 3D-FFT local spectra and multi-dimensional plane projections. The projection scheme divides a 3D-FFT local spectrum into three distinct components, each depicting variations along different dimensions of the data. To detect seismic structures oriented at different angles and to capture directional features within 3D volume, we modify the center-surround model to incorporate directional comparisons around each voxel. The weighted combination of the obtained features then yields a saliency map. Experimental results on a real seismic dataset from the Great South Basin in New Zealand show the effectiveness of the proposed algorithm in the detection of complex fault networks, which are hardly conspicuous within original seismic volume. The subjective evaluation of the results show that the proposed method outperforms the state-of-the-art saliency algorithms and seismic attributes in detecting complex structures and holds a promising future in computer-aided extraction of other geologic features as well.

This paper develops a new framework for data compression in seismic sensor networks by using the distributed principal component analysis (DPCA). The proposed DPCA scheme compresses all seismic traces in the network at the sensor level. First of all, the statistics of the seismic traces acquired at all sensors are represented by a mixture model of a number of probability density functions. Based on this mixture model, the DPCA finds the global PCs at the fusion center. These PCs are then sent back to all sensors so that each sensor projects its own traces over these PCs. This scheme does not require transmitting the original traces, here, leading to a low computational load and a high compression ratio, compared with compression obtained using the local PC analysis (LPCA). Furthermore, we develop an efficient communication solution for the DPCA implementation on practical sensor networks. Finally, the proposed scheme is evaluated using real and synthetic seismic data showing improved performance over the LPCA and the traditional 2-D discrete cosine transform (DCT-2-D) compression. Specifically, to preserve a given signal energy during the compression, the DPCA is shown to achieve a higher compression ratio than the LPCA and the DCT-2-D.

Phase Congruency (PC) can highlight small discontinuities in images with varying illumination and contrast using the congruency of phase in Fourier components. PC can not only detect the subtle variations in the image intensity but can also highlight the anomalous values to develop a deeper understanding of the images content and context. In this paper, we propose a new method based on PC for computational seismic interpretation with an application to subsurface structures delineation within migrated seismic volumes. We show the effectiveness of the proposed method as compared to the edge- and texture-based methods for salt domes boundary detection. The subjective and objective evaluation of the experimental results on the real seismic dataset from the North Sea, F3 block show that the proposed method is not only computationally very efficient but also outperforms the state of the art methods for salt dome delineation.

In this letter, we present a novel supervised codebook-based learning model for salt-dome detection in seismic imaging using texture-based attributes. The proposed algorithm is data driven and overcomes the limitations of existing texture-attributes-based salt-dome detection techniques which are heavily dependent upon the relevance of attributes to the geological nature of salt domes and the number of attributes used for classification. The algorithm works by combining the attributes from the gray-level cooccurrence matrix (GLCM) and those from the Gabor filter, with a codebook-based learning approach to delineate salt boundaries in seismic data. The combination of GLCM- and Gabor-filter-based attributes ensures that the algorithm works well even in the absence of strong reflectors along the salt boundary. Contrary to existing salt-dome detection techniques, our algorithm works with a codebook of small size and is shown to be robust and computationally efficient. The learning properties of the codebook-based model make the algorithm flexible and adaptable to the nature of time-scale varying data acquired in seismic surveys. We used the Netherlands F3 block to evaluate the performance of the proposed algorithm. Our experimental results show that the proposed codebook-based workflow can detect salt domes with good accuracy, superior to existing salt-dome detection techniques.

Salt bodies play an important role in subsurface geology, therefore accurate salt dome detection is essential for any seismic interpretation task. Detecting salt body boundary and shape accurately, however, is very difficult due to large noise and amplitude variations in seismic data. Due to the limitations of manual picking, automatic segmentation algorithms are preferred to locate salt domes within the seismic images. In this work, we propose a robust salt dome detection technique based on combined edge and texture attributes. The proposed algorithm overcomes the drawbacks of existing texture attributes based salt dome detection techniques which are heavily dependent upon the relevance of attributes to the geological nature of salt domes, and the number of attributes used for classification. The combination of edge based and texture based attributes ensures that the proposed algorithm works well even if the salt boundary is represented only by a weak reflector. We tested the proposed algorithm on the Netherlands offshore F3 block. Our experimental results show that the proposed algorithm can detect salt boundaries with high accuracy superior to existing gradient based as well as texture based techniques when used separately.

Rapid growth of emerging medical applications such as e-health and tele-medicine requires fast, low cost, and often lossless access to massive amount of medical images and data over bandlimited channels. In this paper, we first show that significant amount of correlation and redundancy exist across different medical images. Such a correlation can be utilized to achieve better compression, and consequently less storage and less communication overhead on the network. We propose a novel memory-assisted compression technique, as a learning-based universal coding, which can be used to complement any existing algorithm to further eliminate redundancies across images. The approach is motivated by the fact that, often in medical applications, massive amount of correlated images from the same family are available as training data for learning the dependencies and deriving appropriate reference models. Such models can then be used for compression of any new image from the same family. In particular, Principal Component Analysis (PCA) is applied on a set of images from training data to form the required reference models. The proposed memory-assisted compression allows each image to be processed independently of other images, and hence allows individual image access and transmission. Experimental results on X-ray images show that the proposed algorithm achieves 20% improvement over and above traditional lossless image compression methods reported in the literature.

The identification of reservoir regions has a close relationship with the detection of faults in seismic volumes. However, only relying on human intervention, most fault detection algorithms are inefficient. In this paper, we present a new technique that automatically tracks faults across a 3D seismic volume. To implement automation, we propose a two-way fault line projection based on estimated tracking vectors. In the tracking process, projected fault lines are integrated into a synthesized line as the tracked fault line, through an optimization process with local geological constraints. The tracking algorithm is evaluated using real-world seismic data sets with promising results. The proposed method provides comparable accuracy to the detection of faults explicitly in every seismic section, and it also reduces computational complexity.

Sign language continues to be the preferred method of communication among the deaf and the hearing-impaired. Advances in information technology have prompted the development of systems that can facilitate automatic translation between sign language and spoken language. More recently, systems translating between Arabic sign and spoken language have become popular. This paper reviews systems and methods for the automatic recognition of Arabic sign language. Additionally, this paper highlights the main challenges characterizing Arabic sign language as well as potential future research directions.

We present in this paper a new global Full-Reference (FR) image quality metric (IQM) based on the fusion of several conventional FR metrics using an ANN learning algorithm. The fusion is shown to result in improved performance compared to individual FR metrics. Indeed, existing FR metrics can provide excellent results for specific degradations but poor results for others. Here, we propose to overcome this limitation by first improving the performance of existing FR metrics across different degradations through a ranking process. Then, using an Artificial Neural Network, we fuse the best-performing measures into a single metric called Global Index Quality Metric (G-IQM). The experimental results using the TID 2008 image database demonstrate that this new G-IQM metric achieves consistent image quality evaluation results with subjective evaluation.

A plethora of Image Quality Metrics (IQM) has been proposed during the last two decades. However, at present time, there is no accepted IQM able to predict the perceptual level of image degradation across different types of visual distortions. Some measures are more adapted for a set of degradations but inefficient for others. Indeed, the efficiency of any IQM has been shown to depend upon the type of degradation. Thus, we propose here a new approach for predicting the type of degradation before using IQMs. The basic idea is first to identify the type of distortion using a Bayesian approach, then select the most appropriate IQM for estimating image quality for that specific type of distortion. The performance of the proposed method is evaluated in terms of classification accuracy across different types of degradations.

In this paper, we present a novel audio coder using the discrete wavelet transform (DWT) and warped linear prediction (WLP). In contrast to conventional LP, WLP allows for the control of frequency resolution to closely match the response of the human auditory system. The structure of the system is similar to the transform coded excitation techniques used in wideband speech coding, where LP has been replaced with WLP, and the residual is analyzed by a wavelet filterbank designed to approximate the critical bands. The inherent shaping of the WLP synthesis filter, and a controlled bit allocation to the wavelet coefficients helps minimise the perceptually significant noise due to the quantization error in the residual. For monophonic signals sampled at 44.1 kHz, the coder achieves near transparent to transparent quality for a variety of speech and music signals at an average bitrate of about 64 kb/s. Tests also show that the coder (in its initial implementation) delivers superior quality to the MPEG layer III and comparable quality to the MPEG2-AAC codec when operating at the same bitrate

We propose in this paper a model based technique for the detection of human faces from rich still color images. Traditionally, color images are represented in the RGB color space. RGB space, however, is not only a 3-dimensional space but also includes brightness or luminance which is not a reliable criterion for skin separation. To avoid the effect of luminance, we propose to work in the chromatic or pure color space. Using such space, a Gaussian model for the skin color pixels is developed and a skin likelihood image is obtained. Such image is then transformed into a binary image using adaptive thresholding. Finally, bright regions satisfying certain "facial" properties are obtained followed by a template matching stage. The method presented here is shown to provide robust detection under different environments and found to achieve very satisfactory results when compared to traditional "mug shot" based approaches.

In this letter, a new multimodulus algorithm for blind equalization of complex communication channels is derived by solving a constrained optimization problem with relaxation. The intersymbol interference (ISI) optimization and phase-recovery capabilities of the proposed algorithm are analyzed. It is shown from computer simulations that superior performance for the derived algorithm over Lin's algorithm is obtained.

In this work, a new algorithm, based on the minimum-disturbance principle with relaxation, is presented for the blind equalization of complex signals. This algorithm combines the benefits of the well-known reduced constellation algorithm (RCA) and constant modulus algorithm (CMA). The convergence characteristics of the proposed algorithm are demonstrated by way of simulations. In addition, closed form expressions are obtained for the statistical (dispersion) constants used in these algorithms.

In this paper, we present a novel bitstream scalable audio coder. In the proposed coder, the full bandwidth of input audio is first split into two. A hybrid WLPC-wavelet representation is used to encode the low frequency components (<11 kHz). In this method, the excitation to the WLPC synthesis filter is decomposed into subbands using a wavelet filterbank, and perceptually encoded. Two stage quantisation of the wavelet coefficients is used to provide scalability. The high frequency components of the input are assumed to be noisy, and efficiently encoded using an LPC noise model. The output bitstream is capable of being decoded at rates between 16 kbit/s and 80 kbit/s. As the bitrate increases, so too does the signal quality. At 80 kbit/s, the quality is near transparent. At the intermediate rates, the coder gives comparable performance to the MPEG layer III coder, when the MPEG coder operates at similar, but fixed, bitrates.

Electroencephalogram (EEG) signals have long been used for the analysis of brain activities and for the detection of abnormalities (such as seizures). More recently, and with advance of computer technology, we have seen new applications using EEG signals in the control of PC keyboards through BCIs (Brain Computer Interfaces). These EEG signals are normally collected through multi-sensors (8,12, or 16 channels). For proper interpretation of such data, several techniques have been proposed to extract features from the collected multi-channel data, then analyse them, or classify them into patterns. However, most existing techniques do not take into consideration the inherent relationship among features across channels. Here, we propose a scheme based on a hybrid information maximization concept (HIM) to process multi-channel data for optimal feature extraction. The experiments carried show a clear advantage of the approach over principal component and canonical correlation analysis.

In this paper, we present results for a wavelet transform (WT) excited warped linear prediction (WLP) audio coder. In contrast to conventional LP, WLP allows for the control of frequency resolution to closely match the response of the human auditory system. The structure of the system is similar to the transform-coded excitation techniques used in wideband speech coding, where LP has been replaced with WLP. Quantisation of the wavelet coefficients is aided by a psychoacoustic model to minimise the perceptually significant noise due to quantisation error. For monophonic signals sampled at 44.1 kHz, the coder achieves near transparent quality for a variety of speech and music signals at an average bit-rate of 64 kb/s. When compared to MPEG layer III at the same bit-rate, the coder delivers superior quality. The power of the proposed coder resides in its easy scalability to lower bitrates.

This paper presents a new classifier combination technique based on the Dempster-Shafer theory of evidence. The Dempster-Shafer theory of evidence is a powerful method for combining measures of evidence from different classifiers. However, since each of the available methods that estimates the evidence of classifiers has its own limitations, we propose here a new implementation which adapts to training data so that the overall mean square error is minimized. The proposed technique is shown to outperform most available classifier combination methods when tested on three different classification problems.

A novel compression algorithm for fingerprint images is introduced. Using wavelet packets and lattice vector quantization , a new vector quantization scheme based on an accurate model for the distribution of the wavelet coefficients is presented. The model is based on the generalized Gaussian distribution. We also discuss a new method for determining the largest radius of the lattice used and its scaling factor , for both uniform and piecewise-uniform pyramidal lattices. The proposed algorithms aim at achieving the best rate-distortion function by adapting to the characteristics of the subimages. In the proposed optimization algorithm, no assumptions about the lattice parameters are made, and no training and multi-quantizing are required. We also show that the wedge region problem encountered with sharply distributed random sources is resolved in the proposed algorithm. The proposed algorithms adapt to variability in input images and to specified bit rates. Compared to other available image compression algorithms, the proposed algorithms result in higher quality reconstructed images for identical bit rates.

A technique named least squares harmonic (LSH) is proposed for speech decomposition. The problem of harmonic estimation for speech is formulated as a solution to two sets of linear equations derived from minimising the mean squared error between original and estimated signals. The algorithm assumes that a good initial estimate of the pitch period is available. The performance of the algorithm is comparable to that of the total least square Prony method (TLSP) at high signal-to-noise ratios, however, at very low SNR, the proposed algorithm leads to a much more accurate harmonic representation. The approach used here has a great potential in coding and recognition applications.

An EEG feature selection technique for the purpose of classification is developed. The technique selects those features that have maximum mutual information with the specified classes of interest (two classes in this case). Obviously, the simplest way is to consider all possible feature subsets (M out of N). However, even with a small number of features, this procedure is computationally impossible and can not be used in practice. Given the fact that most features used to represent the EEG signal are sets of features (such as AR parameters), our technique considers a trade off between computational cost and chosen feature combination. This contrasts other techniques which select features individually. The classification accuracy of features obtained by applying our technique outperforms those obtained by applying individual feature selection methods when applied to EEG signals.

A novel feature selection algorithm is derived for multi-channel data. This algorithm is a hybrid information maximisation (HIM) technique based on (1) maximising the mutual information between the input and output of a network using the infomax algorithm proposed by Linsker (1988), and (2) maximising the mutual information between outputs of different network modules using the Imax algorithm introduced by Becker (see Network Computation in Neural Systems, vol.7, p.7-31, 1996). The infomax algorithm is useful in reducing the redundancy in the output units, while the Imax algorithm is capable of selecting higher order features from the input units. In this paper, we analyse the two methods and generalise the learning procedure of the Imax algorithm to make it suitable for maximising the mutual information between multi-dimensional output units from different network modules contrary to the original Imax algorithm which only maximises mutual information between two output units. We show that the proposed HIM algorithm provides a better representation of the input compared to the original two algorithms when used separately. Finally, the HIM is evaluated with respect to biological plausibility in the case of feature selection from two-channel EEG data.

The introduction of imaging and anatomical models can be used to develop robust algorithms for mammographic image analysis. The key to the proposed technique is to recognise that at any site in the observed image, a combination of tissues is present. The relationship between the different tissues is represented by a statistical model which is dictated by the imaging system. Markov random fields are used to model the anatomical knowledge. The two models are combined into a Bayesian framework to segment the image and extract regions of interest. Results indicate that reliable and verifiable analysis techniques can be developed utilising physically justified models. Representing the mammographic images in the proposed framework is intended to be more suitable for interpretation by human specialists.

A novel compression algorithm for fingerprint images is introduced. Using wavelet packets and lattice vector quantisation, a new vector quantisation scheme based on an accurate model for the distribution of the wavelet coefficients is presented. In the new algorithm, no assumptions are made about the lattice parameters and no training and multi-quantising are required. The proposed algorithms achieve the best rate-distortion performance by adapting to the statistical characteristics of the source image in each sub-image. Compared to other available image compression algorithms, the proposed algorithms result in higher quality reconstructed images for identical bit rates.

This paper examines a new method for coding high quality digital audio signals based on a combination of linear predictive coding (LPC) and the discrete wavelet transform (DWT). In this method, a linear predictor is first used to model each audio frame. Then, the prediction error is analyzed using the DWT. The LPC coefficients and DWT coefficients are quantized using a novel bit allocation scheme which minimizes the overall quantization error with respect to the masking threshold. The proposed coder is capable of delivering near-transparent audio signal quality at encoding bit rates of around 90-96 kb/s. Objective and subjective results suggest that the proposed coder operating at 90-96 kb/s has a performance comparable to that of the MPEG layer II codec operating at 128 kb/s.

A new method for two-band approximation of excitation signals in an LPC model, to improve speech naturalness in very low rate coding, is proposed. Based on a simplified model of multi-band excitation, the method accurately determines the degree of periodicity, using the concept of instantaneous frequency (IF) estimation in the frequency domain. The harmonic structure in the spectrum of LPC residual, within individual bands, is identified based on flatness of the IF as a criterion for pitch and voicing detection. On this basis, the excitation is modelled by combining a predefined periodic signal in the lower band and a random signal in the higher band. It is shown that this improves considerably the naturalness of reconstructed speech in very low rate coding in comparison with that obtained using traditional binary excitation. The performance of the technique is also given in temporal decomposition (TD) based coding at 800 b/s.

In this paper, we propose a new combined harmonic-wavelet representation for audio where a harmonic analysis-synthesis scheme is used, first, to approximate each audio frame as a sum of several sinusoids. Then, the difference between the original signal and the reconstructed harmonic signal is analyzed using a wavelet filtering scheme. After each step (harmonic analysis and wavelet filtering), parameters are quantized and encoded. Compared to previously proposed methods, our audio coder uses different harmonic analysis-synthesis and wavelet filtering schemes. We use the total least squares (TLS)-prony algorithm for the harmonic analysis-scheme, and an M-band wavelet transform for analyzing the residual. Altogether, our proposed coder is capable of delivering excellent audio signal quality at encoder bitrates of 60-70 kb/s.

Temporal decomposition (TD) is an effective technique to compress the spectral information of speech through orthogonalization of the matrix of spectral parameters leading to an efficient rate reduction in speech coding applications. The performance of TD is a function of the parameters used. Although "decomposition suitability" of a parameter set is typically defined on the basis of "phonetic relevance" criterion, it can not be directly used in speech coding. Instead, quality evaluation of reconstructed speech is more appropriate. In this paper, we extend our earlier work in this area and attempt to assess several "popular" spectral parameter sets from the viewpoint of decomposition suitability in very low bit-rate speech coding using parametric, perceptually-based spectral, and energy distance measures.

A new compression algorithm for fingerprint images is introduced. Using lattice vector quantization (LVQ), a technique for determining the largest radius of the lattice and its scaling factor is presented. The design is based on obtaining the smallest possible expected total distortion (ETD) measure, using a given bit budget, while using the smallest codebook size. In the proposed piecewise-uniform pyramid LVQ, the wedge problem encountered with the pyramidal lattice point shells is resolved. At very low bit rates, for the coefficients with high-frequency content, the positive-negative mean (PNM) method is proposed to improve the resolution of the reconstructed image. The proposed algorithm results in a high compression ratio and a high reconstructed image quality with a low computational load compared to other existing algorithms.

This paper investigates a hybrid scalar-vector quantization scheme for coding high quality audio signals. A wavelet packet transform (WPT) is used to decompose the audio signal into frequency bands slightly finer than the critical band divisions. A masking model computation is then used as input to the hybrid quantization scheme, where scalar quantization is used for coding the subbands from 0-5.5 kHz, and vector quantization is used for coding the subbands from 5.5-22 kHz. The performance of the proposed coder is assessed from segmental signal-to-noise ratios (SNR) and the perceived quality for a number of signals. The perceived quality is determined from informal comparisons between the uncoded signals at the original bitrate of 705 kb/s, and the same signals coded with (1) the proposed coder at 80 kb/s, (2) a coder using only scalar quantization at both 128 kb/s and 96 kb/s, and (3) the MPEG layer III coder at 64 kb/s. The comparisons indicate that a very good coder quality is possible with the proposed coder at bitrates of approximately 80 kb/s. This represents a saving of about 16 kb/s over full scalar quantization with a similar quality. Further bitrate reduction with the proposed coder is possible by entropy coding of the scalar quantized transform coefficients and the VQ indices.

Temporal decomposition (TD) is a method to reduce the correlation between speech spectral parameter sets using orthogonalization. The resulting parameters (so-called weightings or target vectors) correspond, mostly, to least-dependent phonetic states, which produce the desired sounds. Other phonetic states can be interpreted as transition states which force the event functions to overlap. On the basis of this assumption, we propose to approximate the event functions by sample functions to focus on the phonetically important states, and to use the resulting parameters for constructing a very low-rate speech coder. The objective and subjective evaluation of synthesized speech, mostly from the intelligibility view point, confirm the assumptions made on the events.

A multiscale morphological dilation-erosion smoothing operation and its associated scale-space expansion for multidimensional signals are proposed. Properties of this smoothing operation are developed and, in particular a scale-space monotonic property for signal extrema is demonstrated. Scale-space fingerprints from this approach have advantages over Gaussian scale-space fingerprints in that: they are defined for negative values of the scale parameter; have monotonic properties in two and higher dimensions; do not cause features to be shifted by the smoothing; and allow efficient computation. The application of reduced multiscale dilation-erosion fingerprints to the surface matching of terrain is demonstrated.

Most current work in the area of high quality audio coding falls under one of two categories: transform or sub-band coding. LPC coders since based on modelling human voice production systems are found to be inappropriate in modelling music and other non-speech sounds. A more improved model for such signals is shown to be the multipulse LPC model. In this paper we propose to improve the quality of the multipulse model by first passing the signal of interest through a filter bank and then extracting the multipulse parameters from each of the bandpass filter outputs. The idea of the wavelet decomposition is utilised for the design of the filter bank. Both the multipulse model and the wavelet decomposition are well known. But a combination of both has not been exploited yet. This combination is expected to lead to a new way in high quality low bit rate audio coding.

A technique for representing and recognising 3-D or space curves is presented. In the proposed algorithm, the space curves are represented by a set of two zero-crossing representations which are constructed based on the dyadic wavelet transform. These representations are then described in the form of an ordered set of complex numbers which is referred to as the compact representation of the space curves. A string-matching technique is adapted for comparing two curves using their compact representations. Experimental results show that the proposed technique can be used for recognising space curves under similarity transformation with and without additive noise.

This paper considers the problem of parameter estimation of Gaussian autoregressive (AR) processes in the presence of additive white Gaussian noise. The proposed algorithm is based on formulating the estimation problem as an iterative expectation-maximisation (EM) procedure. The observations are seen as the 'incomplete' data and the set formed by the AR process and the noise process represents the 'complete' data. The algorithm is guaranteed to converge in the likelihood function of the parameters. The algorithm is easily generalised to other structures of the covariance matrix of the additive noise. Performance results show that the algorithm is successful in estimating the parameters even at very low signal-to-noise ratios (SNR).>

We present the theory of multiscale dilation-erosion scale-space and the process of feature extraction via morphological scale-space fingerprints. We then discuss the reduced form of the fingerprints and state the scale-space causality theorem. These fingerprints are then applied to the recognition of multiple objects from range data. The proposed recognition system is invariant to translation, rotation, scale, and partial occlusion. We demonstrate results for the recognition of human faces in a scene, and the recognition of mountain features in a digital elevation map.>

A procedure for detecting step edges in noisy signals that does not involve any prefiltering of the data is proposed. It locates step edges in 1-D noisy signals as follows. First, it computes the eigenvectors corresponding to the three smallest eigenvalues of a matrix formed with the discrete Fourier transform of the given data. Next, it estimates the edge locations by finding the local minima in the sum of the spectra of the computed eigenvectors. The technique computes a point edge map for 2-D images by analyzing each row, column, and 45 degrees and 135 degrees diagonal in the image. The computational complexity of the proposed procedure is determined.

A maximum likelihood (ML) estimator for the parameters of Gaussian versions of the fractionally differenced white noise process is developed. A closed-form expression for the likelihood equation is given, and Cramer-Rao bounds are computed for finite size sample data sets. It is shown how the theory can be extended to the case where the fractionally differenced white noise process is observed in the presence of white noise. The results obtained with this ML approach are satisfactory, with a mean square error which is very close to the theoretically computed Cramer-Rao bound.>