Seong-Won Lee

Optical flow-based real-time object tracking using non-prior training active feature model

This paper presents a feature-based object tracking algorithm using optical flow under the non-prior training (NPT) active feature model (AFM) framework. The proposed tracking procedure can be divided into three steps: (i) localization of an object-of-interest, (ii) prediction and correction of the objects position by utilizing spatio-temporal information, and (iii) restoration of occlusion using NPT-AFM. The proposed algorithm can track both rigid and deformable objects, and is robust against the objects sudden motion because both a feature point and the corresponding motion direction are tracked at the same time. Tracking performance is not degraded even with complicated background because feature points inside an object are completely separated from background. Finally, the AFM enables stable tracking of occluded objects with maximum 60% occlusion. NPT-AFM, which is one of the major contributions of this paper, removes the off-line, preprocessing step for generating a priori training set. The training set used for model fitting can be updated at each frame to make more robust objects features under occluded situation. The proposed AFM can track deformable, partially occluded objects by using the greatly reduced number of feature points rather than taking entire shapes in the existing shape-based methods. The on-line updating of the training set and reducing the number of feature points can realize a real-time, robust tracking system. Experiments have been performed using several in-house video clips of a static camera including objects such as a robot moving on a floor and people walking both indoor and outdoor. In order to show the performance of the proposed tracking algorithm, some experiments have been performed under noisy and low-contrast environment. For more objective comparison, PETS 2001 and PETS 2002 datasets were also used.

Noise-adaptive spatio-temporal filter for real-time noise removal in low light level images

Noise reduction gradually becomes one of the most important features in consumer cameras. The video signal is easily interfered by noise during acquisition process especially in low light environment. Many of the state-of-the-art filters for noise reduction perform-well for high contrast images. However, for low light images, the filter performance degrades seriously. In this paper, we propose a noise-adaptive spatio-temporal (NAST) filtering for removal of noise in low light level images. The proposed algorithm consists of a statistical domain temporal filter (SDTF) for moving area and a spatial hybrid filter (SHF) for stationary area. By minimizing required resources for implementation, we present a high quality, low-cost noise reduction filter for low light images. Since the proposed algorithm is designed for real-time implementation, it can be used as a pre-filter for a DCT-based encoder to enhance the coding efficiency of many commercial applications such as low cost camcorders, digital cameras, CCTV, and surveillance video systems.

Robot kinematic control based on bidirectional mapping neural network

The authors present a novel method of accomplishing robot kinematic control based on a bidirectional mapping neural network (BMNN). The BMNN constructed is composed of a multilayer feedforward network with hidden units having sinusoidal activation functions and a feedback network forming a recurrent loop around the feedforward network. The feedforward network can be trained to accurately represent the forward kinematic equations of a robot arm. The feedback network iteratively generates joint-angle updates based on a Lyapunov function to modify the current joint angles in such a way that the output of the forward network converges to the desired Cartesian position and orientation. The proposed BMNN offers the following advantages over the conventional approaches: (1) the accurate computation of robot forward and inverse kinematic solutions with simple training; (2) the ability to handle one-too-many inverse mapping required for redundant arm kinematics solutions; and (3) the automatic generation of arm trajectories. Simulation results are shown

A high performance NIDS using FPGA-based regular expression matching

A Network Intrusion Detection System (NIDS) monitors all incoming packets in the network and detects packets that are malicious to the internal system. The NIDS should also have ability to update the detection rules because new attack patterns are unpredictable. Incorporating FPGAs into the NIDS is one of the best solutions that can provide both high performance and high flexibility comparing to the other approaches such as software solutions. In this paper we propose a novel approach to design the parallel comparator of NIDS that can not only minimize additional resources but also maximize the processing performance. The performance and resource tradeoff due to the implementation of the parallel comparator in the prefix sharing is also analyzed.

Genetic algorithm-based watermarking in discrete wavelet transform domain

This paper presents a watermarking algorithm in the discrete wavelet transform domain using evolutionary algorithm. The proposed algorithm consists of wavelet-domain watermark insertion and genetic algorithm-based watermark extraction. More specifically watermark is inserted to the low-frequency region of wavelet transform domain, and watermark extraction is efficiently performed by using the evolutionary algorithm. The proposed watermarking algorithm is robust against various attacks such as JPEG image compression and geometric transformations.

BAYESNET: Bayesian classification network based on biased random competition using Gaussian kernels

A new neural network architecture referred to as BAYESNET (Bayesian network) is presented. BAYESNET is capable of learning the probability density functions (PDFs) of individual pattern classes from a collection of learning samples, and designed for pattern classification based on the Bayesian decision rule. In BAYESNET, the PDF of a class is represented in terms of the sum of Gaussian subclass PDFs with unknown means, covariances and subclass probabilities that are to be determined through learning. The unique feature of learning the PDF of a class in BAYESNET is the random assignment of a sample of a class to subclasses, i.e., a sample is randomly assigned to a particular subclass for learning according to the probability of the sample to belong to individual subclasses. The property of Gaussian function provides efficient learning of parameters. It is shown that the learned parameters agree with those obtained by the maximum likelihood estimation of the sample set.<<ETX>>

Machine acquisition of skills by neural networks

Presents a theory and an architecture on machine skill acquisition and its implementation in neural networks. Particular emphasis is given to the skill acquisition in man/machine systems where the neural network observes control behavior of a human expert and learns rules behind his expertise. The paradigm of machine acquisition of skills implies the machine exploitation of its own skills through the exploration of experience based on the transferred skills. A new neural network architecture and a learning algorithm, referred to as the hierarchically self-organizing learning (HSOL) network, is presented especially for skill learning. The HSOL network is a dynamically competitive or cooperative network with the ability of self-organizing hidden units, and functions as a universal approximator of arbitrary input-output mappings. This network is applied to machine acquisition of game playing skills, and its performance is compared with that of other networks, including backpropagation, fully connected network, bidirectional associative memory, and recurrent network.<<ETX>>

Throttling-Based Resource Management in High Performance Multithreaded Architectures

Up to now, the power problems which could be caused by the huge amount of hardware resources present in modern systems have not been a primary concern. More recently, however, power consumption has begun limiting the number of resources which can be safely integrated into a single package, lest the heat dissipation exceed physical limits (before actual package meltdown). At the same time, new architectural techniques such as simultaneous multithreading (SMT), whose goal it is to efficiently use the resources of a superscalar machine without introducing excessive additional control overhead, have appeared on the scene. In this paper, we present a new resource management scheme which enables an efficient low power mode in SMT architectures. The proposed scheme is based on a modified pipeline throttling technique which introduces a throttling point at the last stage of the processor pipeline in order to reduce power consumption. We demonstrate that resource utilization plays an important role in efficient power management and that our strategy can significantly improve performance in the power-saving mode. Since the proposed resource management scheme tests the processor condition cycle by cycle, we evaluate its performance by setting a target IPC as one sort of immediate power measure. Our analysis shows that an SMT processor with our dynamic resource management scheme can yield significantly higher overall performance

Handwritten numeral recognition based on hierarchically self-organizing learning networks

Proposes a novel approach to tracing, representation, and subsequently recognition of handwritten numerals. The proposed approach extracts the geometrical and topological features of a numeral and, more importantly, provides the temporal (or dynamic) relationship among the strokes using a heuristic-rule-based tracing algorithm capable of generating a typical stroke sequence of a numeral. With the stroke sequence identified, one is able to extract the feature points (called critical points) of each stroke in an order given by the tracing sequence such that both static features, such as geometrical and topological features, and dynamical features, such as the temporal relationship among strokes, the number of strokes, and the direction of starting and ending strokes, can be preserved. Utilizing the temporal relationship among critical points and their corresponding X (or Y) coordinates as inputs and outputs, one can train a new neural network architecture using a supervised learning algorithm, referred to as a hierarchically self-organizing learning network, as a novel approach to handwritten numeral recognition.<<ETX>>

Evolutionary algorithm-based background generation for robust object detection

One of the most fundamental image analysis models is background generation that helps to extract information and features in still images and sequential images. Since conventional approaches generate the background from intensity values of the image affected by illumination, the resulting background is often unsatisfactory. In case of background generation with sequential images, noises and the changes of illumination causes errors in the generated background. In this paper we propose an efficient background generation algorithm based on generic algorithm. The proposed algorithm calculates the suitability of changing regions of sequential images, and then causes evolution to the next generation to obtain a clear background. In the proposed evolutionary algorithm, the chromosome includes edges and intensity values of the images so that the algorithm can effectively exclude incorrect information caused by the change of illumination and generates an image of pure background.