We present very wide dynamic range (WDR) CMOS image sensor that integrates improved high-speed column-parallel cyclic ADCs. The proposed signal readout technique of extremely short accumulation (ESA) enables the dynamic range to be expanded to a very high illumination region. Including the ESA signals, a total of four different accumulation time signals are read out in one frame period using a burst readout technique. To achieve the high-speed and high-quality signal readout required for multiple exposure signals, column parallel high-speed A/D converters are integrated at the upper and lower sides of the pixel arrays. This improved cyclic ADC, with reduced random noise, better linearity and offset deviation, expands the dynamic range in the low illumination region. The resulting dynamic range is maximally 153 dB. The improved 12-bit cyclic ADC has a differential non-linearity (DNL) of ±0.3 LSB.

抄録全体を表示

We investigated parallel image compression circuits suitable for integration in high-speed CMOS image sensors. We compared the coding efficiency and hardware complexity of several image compression algorithms that use 2-D DCTs using simulation and logic synthesis, and found that using 4×4 point 2-D DCT-based coding methods reduced hardware complexity and improved coding efficiency. We developed a parallel processing architecture for on-sensor image compression that use a processing element array and a data-buffering scheme for parallel data-output. We constructed a prototype 256 × 256 pixel high-speed CMOS image sensor chip that integrates 16 image compression-processing elements and uses 0.25-μm CMOS technology. The area of the image compression circuits is 80% of the image array with 15 μm square pixels. The entire chip could be operated at a clock frequency of 53.6 MHz, and high-speed images compressed by a factor of four could be read out at 10,000 fps using a 32-bit parallel bus.

抄録全体を表示

Multiple-camera systems are emerging gradually. In order to let audiences feel comfortable when cameras are switched or when a free viewpoint video generated is shown, colour correction between cameras is necessary. Colour correction has usually two steps; the first step is to adjust camera parameters such as gain, brightness and aperture before capturing, and the second step is to modify captured videos by image processing. This paper deals with the latter step, which does not need a colour pattern board to be prepared, uses SIFT (Scale Invariant Feature Transform) to detect correspondences for occlusion handling, treats RGB channels independently, and transforms the distorted videos with a non-linear function. Our experimental results show that there was a better correlation between the corrected videos than that between captured ones.

抄録全体を表示

A 3D human pose is estimated from a monocular image using a retrieval-combination approach that exploits the broad capability of example-based approaches and the flexibility of parts-based approaches. Instead of storing and searching for similar full-body examples, we adopt a half-body representation (i.e., either upper-body or lower-body) to reduce a large full-body database into a compact half-body database. The database can create millions of poses by valid half-body combinations. For a given query image, half-body candidates are first retrieved from the database by partial-shape matching. Valid half-body combinations of these candidates are selected based on a learned combination constraint, and then the optimal combination(s) is(are) chosen in a coarse-to-fine evaluation method. We show good experimental results for estimating poses with both synthetic and real images. Our approach has less time and space complexities than example-based approaches and ensures more realistic 3D pose estimates than those of parts-based approaches.

抄録全体を表示

This paper proposes an efficient algorithm to find the worst case flush timings for a given program with respect to the number of branch mispredictions. We first give a basic algorithm based on dynamic programming which takes O(N²F) computation time for a program with Nconditional branches and F flush timings. We then show it can be improved to achieve a computation time of approximately O(NF) for practical programs with its proof obtained through an evaluation with SPEC CPU95 benchmarks.

抄録全体を表示

We propose a Java library BiXJ for bidirectional XML transformation. A bidirectional transformation generates target XML documents from source XML documents in forward transformations, and updates source documents in backward transformations by reflecting back modifications on target documents. The benefit of using BiXJ is that users can get the corresponding backward transformation automatically just by writing one forward transformation. BiXJ has addressed several limitations of the existing bidirectional transformation languages, and can be used for general purpose XML processing. For example, bidirectional XPath expressions written in BiXJ can be used to locate and extract data from XML documents. To validate the usability and expressiveness of BiXJ, we have bidirectionalized some typical examples of XQuery and XSLT with this library. The results of these experiments are promising.

抄録全体を表示

We propose the Eigen Co-occurrence Matrix (ECM) method, which is a modeling method for tracking the behaviors of an individual, system, or network in terms of event sequences of discrete data. Our method uses the correlation between events in a sequence to extract distinct characteristics. A key idea behind the ECM method is to regard a sequence as a serialized sequence that originally had structural relations and to extract the embedded dependencies of the events. To test its retrieval performance, we applied the ECM method to the problem of anomaly detection in intrusion detection systems. Specifically, we used the method to model a UNIX command sequence and attempted to detect intruders masquerading as valid users. The experimental results reveal that the ECM method offers distinct characteristic models for analyzing event sequences.

抄録全体を表示

The transition predicate abstraction framework developed by Podelski, et al. (2005) captures size relations over state transitions which can be used to show infeasibility of certain program computations. In particular, general liveness properties (i.e., properties of infinite computations) can be verified by reducing the verification problem to one of fair termination and then proving that all (infinite) fair computations are infeasible. We present an extension of the algorithm by Podelski, et al. that can be used to improve the precision of transition predicate abstraction as well as speed up analysis time for programs with well-structured control-flow. The main key is to identify sub-computations that can be evaluated independently of their context. Efficiency is then readily improved by analyzing each sub-computation in turn, thus avoiding to reanalyze the effect of a given sub-computations for different contexts. Further, precision can be improved by using stronger methods for extracting summary information about a given sub-computation. We present two versions of the sub-computation based analysis: one for a non-parallel imperative language with loops and recursive procedures, serving as an introduction, and one for the extension of the non-parallel language to a parallel language with synchronous communication via statically named channels.

抄録全体を表示

The ambient calculus is a process algebra designed for describing mobile processes. It has a layered structure of ambients that enables us to describe not only mobile processes but also the world in which the processes move around such as computer networks and freight systems. When we describe such a system with ambients, however, malicious processes can destroy nodes of the network or alter the construction of the system. Thus, several mobility types for the ambient calculus have been proposed to enable us to give each node desirable characteristics that prevent malicious processes from acting harmfully. Gordon and Cardelli, the originators of the ambient calculus, also defined an equivalence relation for the untyped ambient calculus. Our previous work pointed out that there exist identified processes up to the relation that have different properties, and it refined the relation so that we can discriminate those processes. This paper shows that the original relation and our previous relation are no longer available for the typed ambient calculus and it presents another relation that is suitable.

抄録全体を表示

This paper proposes a novel transport-layer protocol called Sender Initiated Congestion Control (SICC) which provides congestion control for Small Group Multicast (SGM). SGM is an effective solution for carrying a large number of simultaneous small-sized group communications. We provide simulated and experimental results to show that SICC is capable of achieving fast congestion avoidance and significantly increases throughput at receivers while still remaining TCP friendly. In SICC, multiple transmission rates are pre-defined as constant values for a given session, and each rate is associated with a different SICC CLASS. Each CLASS corresponds to an SGM group containing receivers with similar acceptable sending rates. The acceptable sending rate of each receiver is estimated at the sender using TCP-Friendly Rate Control (TFRC) in response to the loss event rate reported by the receiver. Based on receiver's reports, the rate that a sender uses to transmit packets to a receiver can be dynamically changed by assigning the receiver to one of the other SICC CLASS. These are the first reported results for congestion control in SGM.

抄録全体を表示

Data mining across different companies, organizations, online shops, or the likes, called sites, is necessary so as to discover valuable shared patterns, associations, trends, or dependencies in their shared data. Privacy, however, is a concern. In many situations it is required that data mining should be conducted without any privacy being violated. In response to this requirement, this paper proposes an effective distributed privacy-preserving data mining approach called CRDM (Collusion-Resistant Data Mining). CRDM is characterized by its ability to resist the collusion. Unless all sites but the victim collude, privacy of a site cannot be violated. Considering that for such applications that need not so high a level of security, excess security assurance would incur extra costs in communication, an extension scheme is also presented so that communication cost can be restrained while still maintaining a user-specified level of security. Results of both analytical and experimental performance study demonstrate the effectiveness of CRDM.

抄録全体を表示

In this paper, we propose a new approach to emotion recognition. Prosodic features are currently used in most emotion recognition algorithms. However, emotion recognition algorithms using prosodic features are not sufficiently accurate. Therefore, we focused on the phonetic features of speech for emotion recognition. In particular, we describe the effectiveness of Mel-frequency Cepstral Coefficients (MFCCs) as the feature for emotion recognition. We focus on the precise classification of MFCC feature vectors, rather than their dynamic nature over an utterance. To realize such an approach, the proposed algorithm employs multi-template emotion classification of the analysis frames. Experimental evaluations show that the proposed algorithm produces 66.4% recognition accuracy in speaker-independent emotion recognition experiments for four specific emotions. This recognition accuracy is higher than the accuracy obtained by the conventional prosody-based and MFCC-based emotion recognition algorithms, which confirms the potential of the proposed algorithm.

抄録全体を表示

The authors propose a method of producing virtual mezzotint using a physics-based rendering approach. Mezzotint is a traditional copperplate printing technique. An important characteristic is its gradations from black to white. This is acquired in three phases during the plate-making process, i.e., by roughening, scraping, and burnishing. Numerous dots and burrs are created with a rocker in the roughening phase over the entire surface of the copper plate to obtain black areas in the print. The burrs are removed in the second phase with a scraper to yield halftones. The plate surface is finally polished with a burnisher in the last phase to produce the white areas. A method of simulating these phases and physical phenomena is discussed in this paper to make mezzotint enjoyable even for beginners and children. Zigzag-stroke patterns and paraboloidal-dot models are applied to the rocker to simulate the roughening phase. Reducing and smoothing models are applied to the scraper and burnisher to simulate the scraping and burnishing phases. The feasibility of the proposed method is demonstrated by observing and comparing actual and virtual plate surfaces with determined patterns and actual pieces of handcrafted work.

抄録全体を表示

The partially connected nature of mobile and ubiquitous computing environments presents software developers with hard challenges. Mobile code has been suggested as a natural fit for simplifying software development for these environments. However, existing strategies for discovering mobile code assume an underlying fixed, stable network. An alternative approach is required for mobile environments, where network size may be unknown and reliability cannot be guaranteed. This paper introduces AMOS, a mobile object platform augmented with a structured overlay network that provides a fully decentralised approach to the discovery of mobile objects. We demonstrate how this technique has better reliability and scalability properties than existing strategies, with minimal communication overhead. Building upon this novel discovery strategy, we show how load balancing of mobile objects in an AMOS network can be achieved through probabilistic means.

抄録全体を表示

In this paper, we investigate conventional registration implementation, consisting of rotation and translation, to design the most precise registration so as to accurately restore the 3D shape of an object. To achieve the most accurate registration, our registration implementation needs robustness against data noise, or initial pose and position of data. To verify the accuracy of our implemented registration, we compare the registration behavior with the registration behavior of conventional methods, and evaluate the numerical accuracy of transformation parameter obtained by our registration. However, registration by rigid-body transformation is not enough for modeling and shape comparison: registration with deformation is needed. In this paper, we extend our robust registration to simultaneously estimate the shape parameter as well as the rigid-body transformation parameter. This extension method assumes that the deformation is formulated strictly from the deformation mechanism. We additionally introduce the applications of our extension method.

抄録全体を表示

“Modeling from Reality” techniques are making great progress because of the availability of accurate geometric data from three-dimensional digitizers. These techniques contribute to numerous applications in many areas. Among them, one of the most important and comprehensive applications is modeling cultural heritage objects. For a large object, scanning from the air is one of the most efficient methods for obtaining 3D data. We developed a novel 3D measurement system, the Floating Laser Range Sensor (FLRS), in which a range sensor is suspended beneath a balloon. The obtained data, however, have some distortions due to movement of the system during the scanning process. We propose two novel methods to rectify the shape data obtained by the moving range sensor. One method rectifies the data by using image sequences; the other rectifies the data without images. To test these methods, we have conducted a digital archiving project of a large-scale heritage object, in which our algorithms are applied. The results show the effectiveness of our methods. Moreover, both methods are applicable not only to our FLRS, but also to moving range sensors in general.

抄録全体を表示

Ad hoc network which requires no infrastructure equipment such as access points is paid great attention in recent years. In ad hoc network, each host controls its wireless access in a distributed fashion. Generally, carrier sensing is used for detecting channel availability. By using RTS and CTS control frames, MAC sub-layer predicts the channel use and resolves hidden-terminal problem. Broadcast nature of RTS and CTS control frames leads to inefficient spatial reuse. One promising solution for this inefficient spatial reuse is usage of directional antenna. In ad hoc networks, heterogeneous network configuration is general, which means each host is equipped with different network facilities. For antennas, it will be general situation that not all hosts are equipped with directional antenna. This paper evaluates throughput performance of the ad hoc network in this realistic situation and shows how much performance improvement deployment of directional antenna will bring to the network.

抄録全体を表示

The Sensor Cube platform is an ultra-compact, modular and power-aware way of building sensor networks; they are based on a stackable hardware design supported by a Tiny OS based operating environment. The Sensor Cube hardware measures 14 × 14 ×18mm³ and features an integrated coplanar antenna, a design that results in an ultra compact footprint. A core characteristic of the system is that its modular design allows for each of the radio, processor, sensing and power management layers to be interchangeable in a Lego-like manner. Moreover, its low power radio (based on the 2.4GHz Nordic nRF2401 design) and microcontroller (based on the Texas Instruments MSP430) allow for very efficient operation. The Sensor Cube operating and software development environment is derived from Tiny OS, which has been modified to meet the hardware requirements, in particular by introducing a power-aware and reliable ALOHA-type MAC protocol. In this paper we present our experience with the Sensor Cube platform and, in particular, the implications of its ultra-compact design on system performance, specifically as it relates to the characteristics and limitations of the radio unit.

抄録全体を表示

Existing cryptography-based algorithms for multi-party computation problems (MPCP) are either too complex to be used practically or applicable only to the specific applications for which they have been developed. In addition, traditional (non-cryptography-based)algorithms do not provide good privacy protection for MPCPs. This paper presents the security system design, intuitive analysis, and empirical evaluation of an agent server platform for protecting privacy in solving multi-party computation problems with traditional algorithms. We have devised the security policies for an agent server platform and showed how the enforcement of those policies can effectively protect privacy while solving many MPCPs. Experimental analysis shows that the presented agent platform security system can significantly enhance privacy protection while solving the problems with traditional algorithms.

抄録全体を表示

Trace analysis for a security protocol represents every possible run as a trace and analyzes whether any insecure run is reachable. The number of traces will be infinite due to (1) infinitely many sessions of a protocol, (2) infinitely many principals in the network, and (3) infinitely many messages that intruders can generate. This paper presents an on-the-fly model checking method by restricting/abstracting these infinite factors to a finite model. First, we restrict a typed process calculus to avoid recursive operations, so that only finitely many sessions are considered. Next, a bound variable is introduced as an index of a message to represent its intended destination, so that an unbounded number of principals are finitely described. Then, messages in which irrelevant parts are reduced in a protocol are unified to a parametric message based on the type information. We implement the on-the-fly model checking method using Maude, and automatically detect the flaws of several security protocols, such as the NSPK protocol and the Woo-Lam protocol, etc..

抄録全体を表示

This paper studies the situation where multiple IP flows are aggregated over a single wireless channel and an error recovery by retransmissions is performed by Selective-Repeat (SR) ARQ. We propose MQ-PFRS (Multi-QoS Per-Flow Resequencing) ARQ that provides a differentiated service for an IP flow depending on its QoS class, such as real-time or non-real-time. MQ-PFRS ARQ eliminates interferences among IP flows by resequencing received packets independently for each IP flow. It also controls the maximum packet delay by limiting the persistency of retransmissions and the maximum resequencing time for each packet. This paper also presents an analysis of the probability distribution of real-time packet delays. Simulation results show that the delay variation of real-time traffic is effectively controlled by proposed MQ-PFRS ARQ and the packet delay distribution is quite consistent with the results of the analysis. This means that MQ-PFRS is effective for handling multiple QoS classes of IP flows and the quality of real-time traffic transferred by the scheme can be predicted by the analysis.

抄録全体を表示

Several directional MAC protocols for ad hoc networks using directional antennas have been proposed recently. Although directional antennas have great potential such as high spatial reuse and range extension, directional MAC protocols inherently introduce new problems arising from directivity. Deafness is one of the major problems and reduces the performance, caused by a lack of state information from neighbor nodes. This paper presents two directional MAC protocols, DMAC/DA (Directional MAC with Deafness Avoidance) and RI-DMAC (Receiver-Initiated Directional MAC), which handle the deafness problem, and mainly evaluates these protocols through extensive simulation study. DMAC/DA is a proactive handling method for deafness. In DMAC/DA, WTS (Wait To Send) frames are transmitted to notify the on-going communication to potential transmitters that may experience deafness. In this paper, DMAC/DA is enhanced by the next packet notification, called DMAC/DA with NPN, to distinguish transmitters from neighbors. On the other hand, RI-DMAC handles deafness reactively using a combination of sender-initiated and receiver-initiated operations. In RI-DMAC, each node polls a potential deafness node using RTR (Ready To Receive) after the completion of every dialog. The experimental results show that our proposed protocols outperform existing directional MAC protocols in terms of throughput, control overhead and packet drop ratio.

抄録全体を表示

Recent advances in wireless communication technologies have attracted ad hoc networking, which enables mobile users to communicate with each other without any infrastructure. While many ad hoc routing protocols have been proposed, ad hoc communication and its applications are not widespread. The handiest wireless devices, namely, mobile phone terminals, are expected to expedite widespread use of ad hoc communications. More mobile phones have been equipped with short distance wireless interfaces such as wireless LAN and Bluetooth. When ad hoc communication is realized on such mobile phones with these wireless interfaces, ad hoc communications will be available anytime, anywhere, which we believe will become a driving force for ad hoc networking technology. However, although the performance of mobile phones has improved greatly over the past several years, their resources are still limited compared to laptop PCs. In this paper, we first present design principles for mobile ad hoc communications based on cases of assumed use. According to the design principles, we implement ad hoc networking protocols as well as their applications on mobile phones to verify their effectiveness. Based on the results of evaluations, we discuss the performance characteristics and potential application areas.

抄録全体を表示