Wee Kheng Leow

Image and Video Retrieval

We have witnessed a decade of exploding research interest in multimedia content analysis. The goal of content analysis has been to derive automatic methods for high-level description and annotation. In this paper we will summarize the main research topics in this area and state some assumptions that we have been using all along. We will also postulate the main future trends including usage of long term memory, context, dynamic processing, evolvable generalized detectors and user aspects.

Extraction of rules from artificial neural networks for nonlinear regression

Neural networks (NNs) have been successfully applied to solve a variety of application problems including classification and function approximation. They are especially useful as function approximators because they do not require prior knowledge of the input data distribution and they have been shown to be universal approximators. In many applications, it is desirable to extract knowledge that can explain how Me problems are solved by the networks. Most existing approaches have focused on extracting symbolic rules for classification. Few methods have been devised to extract rules from trained NNs for regression. This article presents an approach for extracting rules from trained NNs for regression. Each rule in the extracted rule set corresponds to a subregion of the input space and a linear function involving the relevant input attributes of the data approximates the network output for all data samples in this subregion. Extensive experimental results on 32 benchmark data sets demonstrate the effectiveness of the proposed approach in generating accurate regression rules.

Rain Removal in Video by Combining Temporal and Chromatic Properties

Removal of rain streaks in video is a challenging problem due to the random spatial distribution and fast motion of rain. This paper presents a new rain removal algorithm that incorporates both temporal and chromatic properties of rain in video. The temporal property states that an image pixel is never always covered by rain throughout the entire video. The chromatic property states that the changes of R, G, and B values of rain-affected pixels are approximately the same. By using both properties, the algorithm can detect and remove rain streaks in both stationary and dynamic scenes taken by stationary cameras. To handle videos taken by moving cameras, the video can be stabilized for rain removal, and destabilized to restore camera motion after rain removal. It can handle both light rain and heavy rain conditions. Experimental results show that the algorithm performs better than existing algorithms

FERNN: An Algorithm for Fast Extraction of Rules fromNeural Networks

Before symbolic rules are extracted from a trained neural network, the network is usually pruned so as to obtain more concise rules. Typical pruning algorithms require retraining the network which incurs additional cost. This paper presents FERNN, a fast method for extracting rules from trained neural networks without network retraining. Given a fully connected trained feedforward network with a single hidden layer, FERNN first identifies the relevant hidden units by computing their information gains. For each relevant hidden unit, its activation values is divided into two subintervals such that the information gain is maximized. FERNN finds the set of relevant network connections from the input units to this hidden unit by checking the magnitudes of their weights. The connections with large weights are identified as relevant. Finally, FERNN generates rules that distinguish the two subintervals of the hidden activation values in terms of the network inputs. Experimental results show that the size and the predictive accuracy of the tree generated are comparable to those extracted by another method which prunes and retrains the network.

Autonomic mobile sensor network with self-coordinated task allocation and execution

This paper describes a distributed layered architecture for resource-constrained multirobot cooperation, which is utilized in autonomic mobile sensor network coverage. In the upper layer, a dynamic task allocation scheme self-organizes the robot coalitions to track efficiently across regions. It uses concepts of ant behavior to self-regulate the regional distributions of robots in proportion to that of the moving targets to be tracked in a nonstationary environment. As a result, the adverse effects of task interference between robots are minimized and network coverage is improved. In the lower task execution layer, the robots use self-organizing neural networks to coordinate their target tracking within a region. Both layers employ self-organization techniques, which exhibit autonomic properties such as self-configuring, self-optimizing, self-healing, and self-protecting. Quantitative comparisons with other tracking strategies such as static sensor placements, potential fields, and auction-based negotiation show that our layered approach can provide better coverage, greater robustness to sensor failures, and greater flexibility to respond to environmental changes

A hybrid mobile robot architecture with integrated planning and control

Research in the planning and control of mobile robots has received much attention in the past two decades. Two basic approaches have emerged from these research efforts: deliberative vs.\ reactive. These two approaches can be distinguished by their different usage of sensed data and global knowledge, speed of response, reasoning capability, and complexity of computation. Their strengths are complementary and their weaknesses can be mitigated by combining the two approaches in a hybrid architecture. This paper describes a method for goal-directed, collision-free navigation in unpredictable environments that employs a behavior-based hybrid architecture with asynchronously operating behavioral modules. It differs from existing hybrid architectures in two important ways: (1) the planning module produces a sequence of checkpoints instead of a conventional complete path, and (2) in addition to obstacle avoidance, the reactive module also performs target reaching under the control of a self-organizing neural network. The neural network is trained to perform fine, smooth motor control that moves the robot through the checkpoints. These two aspects facilitate a tight integration between high-level planning and low-level control, which permits real-time performance and easy path modification even when the robot is en route to the goal position.

Feature Combination and Relevance Feedback for 3D Model Retrieval

Retrieval of 3D models have attracted much research interest, and many types of shape features have been proposed. In this paper, we describe a novel approach of combining the feature types for 3D model retrieval and relevance feedback processing.Our approach performs query processing using pre-computed pairwise distances between objects measured according to various feature types. Experimental tests show that this approach performs better than retrieval by individual feature type.

The analysis and applications of adaptive-binning color histograms

Histograms are commonly used in content-based image: retrieval systems to represent the distributions of colors in images. It is a common understanding that histograms that adapt to images can represent their color distributions more efficiently than do histograms with fixed binnings. However, existing systems almost exclusively adopt fixed-binning histograms because, among existing well-known dissimilarity measures, only the computationally expensive Earth Movers Distance (EMD) can compare histograms with different binnings. This paper addresses the issue by defining a new dissimilarity measure that is more reliable than the Euclidean distance and yet computationally less expensive than EMD. Moreover, a mathematically sound definition of mean histogram can be defined for histogram clustering applications. Extensive test results show that adaptive histograms produce the best overall performance, in terms of good accuracy, small number of bins, no empty bin, and efficient computation, compared to existing methods for histogram retrieval, classification, and clustering tasks.

3D model retrieval with morphing-based geometric and topological feature maps

Recent advancement in 3D digitization techniques have prompted the need for 3D object retrieval. Our method of comparing 3D objects for retrieval is based on 3D morphing. It computes, for each 3D object, two spatial feature maps that describe the geometry and topology of the surface patches on the object, while preserving the spatial information of the patches in the maps. The feature maps capture the amount of effort required to morph a 3D object into a canonical sphere, without performing explicit 3D morphing. Fourier transforms of the feature maps are used for object comparison so as to achieve invariant retrieval under arbitrary rotation, reflection, and non-uniform scaling o the objects. Experimental results show that our method of retrieving 3D models is very accurate, achieving a precision of above .086 even at a recall rate of 1.0.

Automatic extraction of femur contours from hip x-ray images

Extraction of bone contours from x-ray images is an important first step in computer analysis of medical images. It is more complex than the segmentation of CT and MR images because the regions delineated by bone contours are highly nonuniform in intensity and texture. Classical segmentation algorithms based on homogeneity criteria are not applicable. This paper presents a model-based approach for automatically extracting femur contours from hip x-ray images. The method works by first detecting prominent features, followed by registration of the model to the x-ray image according to these features. Then the model is refined using active contour algorithm to get the accurate result. Experiments show that this method can extract the contours of femurs with regular shapes, despite variations in size, shape and orientation.