Clay Spence

Convolutive blind separation of non-stationary sources

Acoustic signals recorded simultaneously in a reverberant environment can be described as sums of differently convolved sources. The task of source separation is to identify the multiple channels and possibly to invert those in order to obtain estimates of the underlying sources. We tackle the problem by explicitly exploiting the nonstationarity of the acoustic sources. Changing cross correlations at multiple times give a sufficient set of constraints for the unknown channels. A least squares optimization allows us to estimate a forward model, identifying thus the multipath channel. In the same manner we can find an FIR backward model, which generates well separated model sources. Furthermore, for more than three channels we have sufficient conditions to estimate underlying additive sensor noise powers. We show the good performance in a real room environments and demonstrate the algorithms utility for automatic speech recognition.

Response error correction-a demonstration of improved human-machine performance using real-time EEG monitoring

We describe a brain-computer interface (BCI) system, which uses a set of adaptive linear preprocessing and classification algorithms for single-trial detection of error related negativity (ERN). We use the detected ERN as an estimate of a subjects perceived error during an alternative forced choice visual discrimination task. The detected ERN is used to correct subject errors. Our initial results show average improvement in subject performance of 21% when errors are automatically corrected via the BCI. We are currently investigating the generalization of the overall approach to other tasks and stimulus paradigms.

Convolutive blind source separation based on multiple decorrelation

Acoustic signals recorded simultaneously in a reverberant environment can be described as sums of differently convolved sources. The task of source separation is to identify the multiple channels and possibly to invert those in order to obtain estimates of the underlying sources. We tackle the problem by explicitly exploiting the nonstationarity of the acoustic sources. Changing cross-correlations at multiple times give a sufficient set of constraints for the unknown channels. A least squares optimization allows us to estimate a forward model, identifying thus the multipath channel. In the same manner we can find an FIR backward model, which generates well separated model sources. Under certain conditions we obtain up to 14 dB signal enhancement in a real room environment.

Learning contextual relationships in mammograms using a hierarchical pyramid neural network

This paper describes a pattern recognition architecture, which we term hierarchical pyramid/neural network (HPNN), that learns to exploit image structure at multiple resolutions for detecting clinically significant features in digital/digitized mammograms. The HPNN architecture consists of a hierarchy of neural networks, each network receiving feature inputs at a given scale as well as features constructed by networks lower in the hierarchy. Networks are trained using a novel error function for the supervised learning of image search/detection tasks when the position of the objects to be found is uncertain or ill defined. We have evaluated the HPNNs ability to eliminate false positive (FP) regions of interest generated by the University of Chicagos Computer-aided diagnosis (CAD) systems for microcalcification and mass detection. Results show that the HPNN architecture, trained using the uncertain object position (UOP) error function, reduces the FP rate of a mammographic CAD system by approximately 50% without significant loss in sensitivity. Investigation into the types of FPs that the HPNN eliminates suggests that the pattern recognizer is automatically learning and exploiting contextual information. Clinical utility is demonstrated through the evaluation of an integrated system in a clinical reader study. We conclude that the HPNN architecture learns contextual relationships between features at multiple scales and integrates these features for detecting microcalcifications; and breast masses.

Detection, synthesis and compression in mammographic image analysis with a hierarchical image probability model

We develop a probability model over image spaces and demonstrate its broad utility in mammographic image analysis. The model employs a pyramid representation to factor images across scale and a tree-structured set of hidden variables to capture long-range spatial dependencies. This factoring makes the computation of the density functions local and tractable. The result is a hierarchical mixture of conditional probabilities, similar to a hidden Markov model on a tree. The model parameters are found with maximum likelihood estimation using the EM algorithm. The utility of the model is demonstrated for three applications; 1) detection of mammographic masses in computer-aided diagnosis 2) qualitative assessment of model structure through mammographic synthesis and 3) compression of mammographic regions of interest.

On-line Convolutive Blind Source Separation of Non-Stationary Signals

We have shown previously that non-stationary signals recorded in a static multi-path environment can often be recovered by simultaneously decorrelating varying second order statistics. As typical sources are often moving, however, the multi-path channel is not static. We present here an on-line gradient algorithm with adaptive step size in the frequency domain based on second derivatives, which we refer to as multiple adaptive decorrelation (MAD). We compared the separation performance of the proposed algorithm to its off-line counterpart and to another decorrelation based on-line algorithm.

Varying complexity in tree-structured image distribution models

Probabilistic models of image statistics underlie many approaches in image analysis and processing. An important class of such models have variables whose dependency graph is a tree. If the hidden variables take values on a finite set, most computations with the model can be performed exactly, including the likelihood calculation, training with the EM algorithm, etc. Crouse et al. developed one such model, the hidden Markov tree ( HMT). They took particular care to limit the complexity of their model. We argue that it is beneficial to allow more complex tree-structured models, describe the use of information theoretic penalties to choose the model complexity, and present experimental results to support these proposals. For these experiments, we use what we call the hierarchical image probability (HIP) model. The differences between the HIP and the HMT models include the use of multivariate Gaussians to model the distributions of local vectors of wavelet coefficients and the use of different numbers of hidden states at each resolution. We demonstrate the broad utility of image distributions by applying the HIP model to classification, synthesis, and compression, across a variety of image types, namely, electrooptical, synthetic aperture radar, and mammograms (digitized X-rays). In all cases, we compare with the HMT.

A Multi-scale Probabilistic Network Model for Detection, Synthesis and Compression in Mammographic Image Analysis

We develop a probabilistic network model over image spaces and demonstrate its broad utility in mammographic image analysis, particularly with respect to computer-aided diagnosis. The model employs a multi-scale pyramid decomposition to factor images across scale and a network of tree-structured hidden variables to capture long-range spatial dependencies. This factoring makes the computation of the density functions local and tractable. The result is a hierarchical mixture of conditional probabilities, similar to a hidden Markov model on a tree. The model parameters are found with maximum likelihood estimation using the expectation-maximization algorithm. The utility of the model is demonstrated for three applications: (1) detection of mammographic masses for computer-aided diagnosis; (2) qualitative assessment of model structure through mammographic synthesis; and (3) compression of mammographic regions of interest.

Integrating neural networks with image pyramids to learn target context

The utility of combining neural networks with pyramid representations for target detection in aerial imagery is explored. First, it is shown that a neural network constructed using relatively simple pyramid features is a more effective detector, in terms of its sensitivity, than a network which utilizes more complex object-tuned features. Next, an architecture that supports coarse-to-fine search, context learning and data fusion is tested. The accuracy of this architecture is comparable to a more computationally expensive non-hierarchical neural network architecture, and is more accurate than a comparable conventional approach using a Fisher discriminant. Contextual relationships derived both from low-resolution imagery and supplemental data can be learned and used to improve the accuracy of detection. Such neural network/pyramid target detectors should be useful components in both user assisted search and fully automatic target recognition and monitoring systems.

Multisensor integration in biological systems

The authors discuss the integration of sensory information in biological systems. In particular, they consider the structure in vertebrate animals that utilizes multiple sensory inputs to orient the sensor platform, i.e. the body or the head, toward objects of interest. This structure is known as the optic tectum in lower vertebrates and the superior colliculus in mammals. The representation of the various sensory modalities on the tectum follows the maplike image format of the retina. This requires in some cases a considerable transformation from the original representation of the sensory input available from the other sensors. As an example, the authors present a detailed discussion of the visual/acoustic object localization system of the barn owl along with a model for the adaptive coregistration of the coordinate systems of the visual and acoustic maps on the tectum.<<ETX>>