The capacity of an order-d associative memory model is O(Nd/logN). The exponential growth of the capacity with respect to d gives higher order models (d > 1) significant advantage over the Hopfield network, whose capacity is limited to O(N/logN). One problem with the higher order models is that their data processing can not be implemented on network models that are widely considered biologically feasible. In this paper, we propose a new higher order associative memory model that can be implemented with a collection of a simple McCulloch-Pitts neuron model with one extension: the synapses or network weights being adaptive or plastic. The plasticity is achieved by allowing the synaptic weights to change along the projection of the current network state to a correlation tensor. We show that our model guarantees convergence with a weak constraint, retains the capacity and basin of attraction of the higher correlation model, and is more robust against noise than the higher correlation model. Extensive numerical experiments agree with our claim. We also propose a d = 2 model with an approximate correlation tensor that brings reduction of computational cost during the retrieval phase. Numerical experiments show that the cost can be reduced by half without affecting the retrieval performance significantly.

The capacity of an order-d associative memory model is O(Nd/logN) where N is the memory size in bit. The exponential growth of the capacity with respect to d gives higher order models (d > 1) a significant advantage over the Hopfield network, whose capacity is limited to O(N/logN). In particular, a second order correlation memory (d = 2) has attractive properties: a small implementation cost of O(N2), a small number of spurious states, and the presence of a diagonalization form. Due to these properties, it is of both practical and scientific interests to investigate biological feasibility of such network. One disadvantage of higher order associative memory is that it cannot be implemented with simple threshold neurons or McCulloch-Pitts neurons, thus a direct implementation of its computational mechanism on a biological substrate is questionable. In this paper, we propose two approximation models of a second order associative memory using threshold logics. Both are two-layered and employ eigenvalue decomposition of the correlation tensor. The first model uses a winner-take-all mechanism and the second uses a multiple selection mechanism. Extensive numerical simulations demonstrate effectiveness of the proposed models.

We present a system for automatic hot spots detection and segmentation in whole body FDG-PET images. The main contribution of our system is threefold. First, it has a novel body-section labeling module based on spatial hidden-Markov models (HMM); this allows different processing policies to be applied in different body sections. Second, the competition diffusion (CD) segmentation algorithm, which takes into account body-section information, converts the binary thresholding results to probabilistic interpretation and detects hot-spot region candidates. Third, a recursive intensity mode-seeking algorithm finds hot spot centers efficiently, and given these centers, a clinically meaningful protocol is proposed to accurately quantify hot spot volumes. Experimental results show that our system works robustly despite the large variations in clinical PET images.

This paper describes a novel classification method for computer aided detection (CAD) that identifies structures of interest from medical images. CAD problems are challenging largely due to the following three characteristics. Typical CAD training data sets are large and extremely unbalanced between positive and negative classes. When searching for descriptive features, researchers often deploy a large set of experimental features, which consequently introduces irrelevant and redundant features. Finally, a CAD system has to satisfy stringent real-time requirements.This work is distinguished by three key contributions. The first is a cascade classification approach which is able to tackle all the above difficulties in a unified framework by employing an asymmetric cascade of sparse classifiers each trained to achieve high detection sensitivity and satisfactory false positive rates. The second is the incorporation of feature computational costs in a linear program formulation that allows the feature selection process to take into account different evaluation costs of various features. The third is a boosting algorithm derived from column generation optimization to effectively solve the proposed cascade linear programs.We apply the proposed approach to the problem of detecting lung nodules from helical multi-slice CT images. Our approach demonstrates superior performance in comparison against support vector machines, linear discriminant analysis and cascade AdaBoost. Especially, the resulting detection system is significantly sped up with our approach.

We present ratio contour, a novel graph-based method for extracting salient closed boundaries from noisy images. This method operates on a set of boundary fragments that are produced by edge detection. Boundary extraction identifies a subset of these fragments and connects them sequentially to form a closed boundary with the largest saliency. We encode the Gestalt laws of proximity and continuity in a novel boundary-saliency measure based on the relative gap length and average curvature when connecting fragments to form a closed boundary. This new measure attempts to remove a possible bias toward short boundaries. We present a polynomial-time algorithm for finding the most-salient closed boundary. We also present supplementary preprocessing steps that facilitate the application of ratio contour to real images. We compare ratio contour to two closely related methods for extracting closed boundaries: Elder and Zucker's method based on the shortest-path algorithm and Williams and Thornber's method based on spectral analysis and a strongly-connected-components algorithm. This comparison involves both theoretic analysis and experimental evaluation on both synthesized data and real images.

This paper presents a novel graph-theoretic approach, named ratio con- tour, to extract perceptually salient boundaries from a set of noisy bound- ary fragments detected in real images. The boundary saliency is defined using the Gestalt laws of closure, proximity, and continuity. This pa- per first constructs an undirected graph with two different sets of edges: solid edges and dashed edges. The weights of solid and dashed edges measure the local saliency in and between boundary fragments, respec- tively. Then the most salient boundary is detected by searching for an optimal cycle in this graph with minimum average weight. The proposed approach guarantees the global optimality without introducing any biases related to region area or boundary length. We collect a variety of images for testing the proposed approach with encouraging results.

Often costs or hazardous conditions prohibit geophysical measurements within the survey area. Surface-wave tomography can be a practical method of imaging the near surface shear-wave velocity structure. We demonstrate that this method can resolve meter-size objects. Surface-wave group-velocity travel times along ray-paths through the survey area are determined as a function of frequency by using the "multiple filter" technique of seismic surface-wave analysis. The spatial distribution of group velocities are calculated from the tomographic inversion of the group travel times. Dispersion curves are generated for a selected portion of the survey area and inverted to give shear-wave velocity structure as a function of depth. The buried objects and the disturbance caused by digging are associated with a depression in the shear-wave velocity.

Describes a technique for determining a classification map or a hypothesis selection map using a system of reaction-diffusion equations. The reaction term of the equations is a replicator equation and provides a mutually exclusive solution. The diffusion term exploits the local consistency of the solution. The technique is efficient for problems with a small labeling space compared to annealing based relaxation techniques. The paper gives experimental results of the technique for target classification and stereo correspondence.

This paper reviews our ongoing efforts on designing a real-time vision system. The system consists of a real-time feature extractor and a relaxation network. The filter system is capable of performing multiple 2D non-separable filters, multi-resolution decomposition, and steerable transform. The relaxation network is capable of performing various relaxation, diffusion and minimization operations. The paper shows several vision tasks which can be implemented effectively on the vision system.