A framework is proposed for segmenting image textures by using Gabor filters to detect boundaries between adjacent textured regions. By performing a multi-channel filtering of the input image with a small set of adaptively selected Gabor filters, tuned to underlying textures, feature images are obtained. To reduce the variance of the filter output for better texture boundary detection, a Gaussian post-filter is applied to the Gabor filter response over each channel. Significant local variations in each channel response are detected using a gradient operator, and combined through channel grouping to produce the texture gradient. A subsequent post-processing produces expected texture boundaries. The effectiveness of the proposed technique is demonstrated through experiments on synthetic and natural textures.

In server load balancing where replicated servers are dispersed geographically and accesses from clients are distributed to replicated servers, a way of distributing the accesses from clients to an adequate server plays an important role from the viewpoint of load balancing. In the paper, we propose a new network paradigm for server load balancing using active anycast. In active anycast, an end user only sends its request to group of servers using an anycast address. When this request arrives at an active router, it selects an adequate server from the viewpoint of load balancing and changes the anycast address of a packet to the unicast address of the selected server. The decision which server is the best one from the viewpoint of server load balancing is made by an active router rather than an end user, so active anycast is a network-initiated method. Simulation results show that active anycast can accomplish efficient server load balancing, even when a small part of routers are equipped with active network technology.

In order to increase the flexibility and the extensibility for packet transmission, the active network approach, that makes network nodes programmable, is very promising. Every packet includes a program specifying behavior of the packet at network nodes, and the network nodes only have the functions to execute programs of the packet. In an active network, network congestion management is achieved more flexibly and intelligently than that in the conventional internet. A number of mechanisms for congestion management are easily developed because various network functions such as QoS routing and congestion detection are easily combined in the same network nodes. By executing different routing protocols that use different levels of the link information, a congestion management scheme can be customized by users. This paper shows how QoS routing and congestion detection achieve a congestion management over an active internetwork system.

This paper proposes a λ-ring system that is a wavelength-based self-healing-ring application unlike ordinary fiber-based ones. To design survivable networks of interconnected such self-healing ring systems, a virtual mesh network scheme is used, in which wavelength assignment in virtual links can be considered according to the λ-ring-system or fiber-ring-system applications of the bidirectional wavelength-path switched architecture. Integer-programming-based design problems are then formulated that minimize the total fiber length in these self-healing-ring applications. Numerical examples show that the λ-ring-system application is always superior to 4-fiber and 2-fiber-ring-system applications and 1+1 end-to-end path protection.

Selective attention mechanism, plays an important role in human visual perception, can be investigated by developing an approach to perceiving the multi-meaningful-dotted-pattern in a color blindness plate (CBP). In this Letter, a perception model driven by a simple active vision mechanism is presented for the image segmentation and understanding of a CBP. Experiments show that to understand one meaningful pattern in an image containing multi-meaningful patterns, the active visual search (i.e., pattern attention) is a very useful function.

A packet transmission delay is an important quality characteristic for various applications including real-time and data applications. In particular, it is necessary to investigate not only a whole distribution of the packet transmission delay, but also the tail part of the distribution, in order to detect the packet loss. In this paper, we analyze the characteristics of the tail part of packet delay distributions by statistical analytic approach. Our analytic results show that the Pareto distribution is most appropriate in 95-99.9% region of the cumulative distribution of packet transmission delays. Based on our statistical analysis, we next propose an adaptive playout control algorithm, which is suitable to real-time applications. Numerical examples show that our algorithm provides the stable packet loss ratio independently on traffic fluctuations.

This paper suggests the way to perform the handover by predicting the movement route of the mobile terminal by considering the movement pattern of the user. By considering the fact that the most users has the constant movement pattern, the channels needed for the handover can be reserved, and the required quality of service (QoS) is maintained during handover. The suggested algorithm makes the channel allocation schemes more efficient.

This paper proposes a framework for building a customized comparison-shopping system in which the users are allowed to add their own shopping stores dynamically. In this framework, a shopping agent is implemented with a robust inductive learning method that automatically constructs wrappers for semi-structured online stores. During learning, strong biases assumed in many existing systems are weakened so that the real stores with reasonably complex document structures can be handled.

Matching pursuit is a signal expansion technique whose efficiency for motion compensated residual image has been successfully demonstrated in the MPEG-4 development. However, one of the practical concerns related to applying matching pursuit algorithm to real-time coding of video is its massive computation required for finding atoms. This paper proposes a new fast method based on three properties of basis functions used in the signal expansion. The first one is the symmetry property of the 1-D bases. The second one is that one can preclude many bases that cannot be atom by checking a simple mathematical condition. The last one is the classification property of 2-D bases in a given dictionary. Experimental result shows that our method can perform the same matching pursuit without any image degradation using only about 40% of computational load required by the conventional fast method based on separability of 2-D Gabor dictionary. Furthermore, if negligible quality degradation is allowed, the method can be extended to perform matching pursuit with only about 10% of the computational load required by the conventional fast method. We apply the proposed fast matching pursuit method to scalable coding of video with two layers.

Cache memories are small fast memories used to temporarily hold the contents of main memory that are likely to be referenced by processors so as to reduce instruction and data access time. In study of cache performance, most of previous works have employed simulation-based methods. However, that kind of researches cannot precisely explain the obtained results. Moreover, when a new processor is designed, huge simulations must be performed again with several different parameters. This research classifies cache structures for superscalar processors into four types, and then represents analytical model of instruction fetch process for each cache type considering various kinds of architectural parameters such as the frequency of branch instructions in program, cache miss rate, cache miss penalty, branch misprediction frequency, and branch misprediction penalty, and etc. To prove the correctness of the proposed models, we performed extensive simulations and compared the results with the analytical models. Simulation results showed that the proposed model can estimate the expected instruction fetch rate accurately within 10% error in most cases. This paper shows that the increase of cache misses reduces the instruction fetch rate more severely than that of branch misprediction does. The model is also able to provide exact relationship between cache miss and branch misprediction for the instruction fetch analysis. The proposed model can explain the causes of performance degradation that cannot be uncovered by the simulation method only.

In this paper, multiuser detector (MUD) based on radial basis function (RBF) is proposed and simulated for a multicode DS/CDMA system in an AWGN and a multipath fading channels. The performance of RBF-based MUD is compared with that of many suboptimal multiuser detectors in terms of bit error probability. To obtain simulation results, importance sampling technique is employed. From the simulation results, it is confirmed that the RBF-based MUD outperforms decorrelating detector, and achieves near-optimum performance under various environments. The results in this paper can be applied to design of MUD for a multicode DS/CDMA system.

In this paper, we first propose a new speech enhancement preprocessing algorithm by combining power subtraction method and maximal ratio combining technique, then apply it to both energy-based and statistical model-based VAD algorithm to improve the performance even in low SNR conditions. We also perform extensive computer simulations to demonstrate the performance improvement of the proposed VAD algorithm employing the proposed speech enhancement preprocessing algorithm under various background noise environments.

Various shadows are one of main factors that cause errors in vision based vehicle detection. In this paper, two simple methods, land mark based method and BS & Edge method, are proposed for vehicle detection and shadow rejection. In the experiments, the accuracy of vehicle detection is higher than 98%, during which the shadows arisen from roadside buildings grew considerably. Based on these two methods, vehicle counting, tracking, classification, and speed estimation are achieved so that real-time traffic parameters concerning traffic flow can be extracted to describe the load of each lane.

Magnetic random access memory (MRAM) possesses the attractive properties of non-volatility, radiation hardness, nondestructive readout, low voltage, high access speed, unlimited read and write endurance and high density. MRAM technology is described for the devices using giant magnetoresistance (GMR) and tunneling magnetoresistance (TMR) materials in this paper. The TMR type MRAM architectures using ferromagnetic tunneling junctions (MTJ) are more attractive for mainstream RAM applications than the GMR type, because the signal of the TMR type is larger than that of the GMR type. A MRAM device with an MTJ plus MOS transistor switch architecture, which can provide large signal-to noise ratio, is detailed. A design of the MTJ element is discussed and the requirements for the junction resistance and the TMR needed for the memory device are demonstrated based on the simple signal voltage calculations. The TMR significantly decreases with increasing bias voltage, which leads to the reduction of the signal voltage for the actual MRAM. A ferromagnetic double tunneling junction is proposed for the high density MRAM application demanding large signal voltage, because of the smaller degradation of the TMR for the bias voltage, compared with that of the conventional single junctions. Recent trials of MRAM fabrication are introduced, which demonstrates high-speed access time. Finally, challenges for the higher bit density MRAM above Gb are discussed, and it is noticed that higher signal voltage, lower power consumption for writing and novel cell designs are needed for the achievement.

A high-speed context-free marker controlled and minima imposition-free watershed transform is proposed for efficient multi-object detection and segmentation from a complex background. The context-free markers are extracted from a complex backgrounded multi-object image using a noise tolerant attention operator. These make high speed marker-controlled watershed possible without over-segmentation and region merging. The proposed method presents a marker-constrained labeling that can speed up the segmentation of the marker-controlled watershed transform by eliminating the necessity of the minima imposition. Simulation results show that the proposed method can efficiently detect and segment multiple objects from a complex background while reducing the over-segmentation and computation time.

We have designed a 32-bit RISC microprocessor with 16-/32-bit fixed-point DSP functionality. This processor, called YD-RISC, combines both general-purpose microprocessor and digital signal processor (DSP) functionality using the reduced instruction set computer (RISC) design principles. It has functional units for arithmetic operation, digital signal processing (DSP) and memory access. They operate in parallel in order to remove stall cycles after DSP or load/store instructions, which usually need one or more issue latency cycles in addition to the first issue cycle. High performance was achieved with these parallel functional units while adopting a sophisticated five-stage pipeline structure. The pipelined DSP unit can execute one 32-bit multiply-accumulate (MAC) or 16-bit complex multiply instruction every one or two cycles through two 17-b 17-b multipliers and an operand examination logic circuit. Power-saving techniques such as power-down mode and disabling execution blocks allow low power consumption. In the design of this processor, we use logic synthesis and automatic place-and-route. This top-down approach shortens design time, while a high clock frequency is achieved by refining the processor architecture.

This paper is intended to provide reliable carrier recovery in environments with a very low C/N (carrier-to-noise power ratio). A demodulation scheme using a carrier recovery circuit supported by frame symbols (CRC-PIDS) is proposed. This scheme uses a recovery order of clock, frame, and carrier, which is effective for carrier recovery in a low C/N channel, and enables coherent detection without differential coding. This paper also evaluates the bit-error-rate (BER) performance of the proposed scheme used with a binary PSK signal with a rate-1/3 4-state turbo code. Computer simulation trials show that the BER performance difference between ideal and practical coherent detections is about 0.2 dB, and that carrier recovery is reliable even at a C/N of -4.8 dB.

In an ideal fiber grating having a uniform refractive index modulation, the reflection or the transmission spectrum is symmetric with equal amount of side lobes on both sides of the resonant wavelength of the fiber grating. It is observed that a long-period fiber grating made by a non-uniform UV laser beam through a uniform amplitude mask has an asymmetric transmission spectrum. The asymmetric characteristic is explained with Mach-Zehnder effect in the long-period fiber grating. The non-uniform UV laser beam makes also a non-uniform index modulation along the fiber core. Therefore, a beam coupled to a cladding mode at a section of the grating can be re-coupled to the core mode after passing a certain distance. The re-coupled beam makes Mach-Zehnder-like interference with the un-coupled core mode. However, it is presented that the asymmetric phenomenon can be overcome by scanning the UV laser beam along the fiber over the mask. The beam scanning method is able to suffer the same fluence of the UV laser beam on the fiber. Finally, a linearly chirped long-period fiber grating was made using the non-uniform UV laser beam. Due to the asymmetricity the chirping effect was not clearly observed. It is also presented that the beam scanning method could remove the asymmetric problem and recover the typical spectrum of the linearly chirped fiber grating.

The present paper discusses several promising application areas for optical data links based on high-performance vertical-cavity surface-emitting laser diodes (VCSELs). Both 850 and 980 nm emission wavelength devices realized in the GaAs-AlGaAs or InGaAs-AlGaAs material systems are considered. We show data transmission results of 10 Gb/s signals at 830 nm wavelength over a new high-bandwidth multimode silica fiber of up to 1.6 km length. The same fiber type is employed to demonstrate the first 40 Gb/s transport over 300 m distance by means of a 4-channel coarse wavelength-division multiplexing approach. A first 1 10 linear VCSEL array capable of 10 Gb/s per channel operation is presented for use in next generation parallel optical modules. To improve the singlemode emission characteristics for output power in the 5 mW range we introduce a new device concept incorporating a long monolithic cavity. For low-cost short-distance data links we investigate graded-index polymer optical fibers and report on up to 9 Gb/s transmission over a length of 100 m. Polymer waveguides are also used in an optical layer of a hybrid electrical-optical printed circuit board. Transmitted 10 Gb/s optical data over a prototype board show the potential of this new technology. Finally we present two-dimensional VCSEL arrays for highly parallel data transport on a CMOS chip level. Both 980 and 850 nm bottom emitting devices with modulation capabilities up to 12.5 Gb/s are discussed.

Many methods have been proposed to detect intrusions; for example, the pattern matching method on known intrusion patterns and the statistical approach to detecting deviation from normal activities. We investigated a new method for detecting intrusions based on the number of system calls during a user's network activity on a host machine. This method attempts to separate intrusions from normal activities by using discriminant analysis, a kind of multivariate analysis. We can detect intrusions by analyzing only 11 system calls occurring on a host machine by discriminant analysis with the Mahalanobis' distance, and can also tell whether an unknown sample is an intrusion. Our approach is a lightweight intrusion detection method, given that it requires only 11 system calls for analysis. Moreover, our approach does not require user profiles or a user activity database in order to detect intrusions. This paper explains our new method for the separation of intrusions and normal behavior by discriminant analysis, and describes the classification method by which to identify an unknown behavior.