Similarity search is an important and fundamental problem, and thus widely used in various fields of computer science including multimedia, computer vision, database, information retrieval, etc. Recently, since loitering behavior often leads to abnormal situations, such as pickpocketing and terrorist attacks, its analysis attracts increasing attention from research communities. In this paper, we present AntiLoiter, a loitering discovery system adopting efficient similarity search on surveillance videos. As we know, most of existing systems for loitering analysis, mainly focus on how to detect or identify loiterers by behavior tracking techniques. However, the difficulties of tracking-based methods are known as that their analysis results are heavily influenced by occlusions, overlaps, and shadows. Moreover, tracking-based methods need to track the human appearance continuously. Therefore, existing methods are not readily applied to real-world surveillance cameras due to the appearance discontinuity of criminal loiterers. To solve this problem, we abandon the tracking method, instead, propose AntiLoiter to efficiently discover loiterers based on their frequent appearance patterns in longtime multiple surveillance videos. In AntiLoiter, we propose a novel data structure Luigi that indexes data using only similarity value returned by a corresponding function (e.g., face matching). Luigi is adopted to perform efficient similarity search to realize loitering discovery. We conducted extensive experiments on both synthetic and real surveillance videos to evaluate the efficiency and efficacy of our approach. The experimental results show that our system can find out loitering candidates correctly and outperforms existing method by 100 times in terms of runtime.

With the proliferation of hand-held devices in recent years, mobile video streaming has become an extremely popular application. However, Internet video streaming to mobile devices faces several problems, such as unstable connections, long latency, high jitter, etc. We present a system, OptVid, which enhances the user's experiences of video streaming service on cellular networks. OptVid takes the user's profile and provides seamless adaptive bitrate streaming by leveraging the video transcoding solution. It provides very agile bitrate adaptation, especially in the mobile scenario where the wireless channel is not stable. We prototype video transcoding on a WiMAX testbed to bridge the gap between the wireless channel capacity and the video quality. Our evaluations reveal that OptVid provides better user experience than conventional schemes in terms of PSNR, video stalls, and buffering time. OptVid does not require any additional storage since it transcodes videos on-the-fly upon receiving requests and delivers them directly to the client.

Automatic detection of immunoreactive areas in fluorescence microscopic images is becoming a key technique in the field of biology including neuroscience, although it is still challenging because of several reasons such as low signal-to-noise ratio and contrast variation within an image. In this study, we developed a new algorithm that exhaustively detects co-localized areas in multi-channel fluorescence images, where shapes of target objects may differ among channels. Different adaptive binarization thresholds for different local regions in different channels are introduced and the condition of each segment is assessed to recognize the target objects. The proposed method was applied to detect immunoreactive spots that labeled membrane receptors on dendritic spines of mouse cerebellar Purkinje cells. Our method achieved the best detection performance over five pre-existing methods.

This paper aims to investigate the performance improvements made possible by combining various major language model (LM) technologies together and to reveal the interactions between LM technologies in spontaneous automatic speech recognition tasks. While it is clear that recent practical LMs have several problems, isolated use of major LM technologies does not appear to offer sufficient performance. In consideration of this fact, combining various LM technologies has been also examined. However, previous works only focused on modeling technologies with limited text resources, and did not consider other important technologies in practical language modeling, i.e., use of external text resources and unsupervised adaptation. This paper, therefore, employs not only manual transcriptions of target speech recognition tasks but also external text resources. In addition, unsupervised LM adaptation based on multi-pass decoding is also added to the combination. We divide LM technologies into three categories and employ key ones including recurrent neural network LMs or discriminative LMs. Our experiments show the effectiveness of combining various LM technologies in not only in-domain tasks, the subject of our previous work, but also out-of-domain tasks. Furthermore, we also reveal the relationships between the technologies in both tasks.

Logistic regression is a powerful machine learning tool to classify data. When dealing with sensitive or private data, cares are necessary. In this paper, we propose a secure system for privacy-protecting both the training and predicting data in logistic regression via homomorphic encryption. Perhaps surprisingly, despite the non-polynomial tasks of training and predicting in logistic regression, we show that only additively homomorphic encryption is needed to build our system. Indeed, we instantiate our system with Paillier, LWE-based, and ring-LWE-based encryption schemes, highlighting the merits and demerits of each instantiation. Besides examining the costs of computation and communication, we carefully test our system over real datasets to demonstrate its utility.

A method based on score level fusion using logistic regression has been developed that uses packet header information to classify Internet applications. Applications are classified not on the basis of the individual flows for each type of application but on the basis of all the flows for each type of application, i.e., the “overall traffic flow.” The overall traffic flow is divided into equal time slots, and the applications are classified using statistical information obtained for each time slot. Evaluation using overall traffic flow generated by five types of applications showed that its true and false positive rates are better than those of methods using feature level fusion.

Effects of electron beam irradiation at 15 keV on graphene are investigated by optical and electron characterization using Raman and two-terminal resistance measurement and photoconductivity measurement. In Raman spectra, increase of defects in D-peak to G-peak ratio by increase of electron irradiation by 70 mC/cm2 was found. Resistance of graphene showed an increase after the irradiation. Rather sensitive change was found in photoconductivity of irradiated graphene under ultra-violet (UV) illumination, suggesting irradiation induced defects affect a photoconductivity properties of the graphene.

This paper reports an assessment of the feasibility and the practicality of a creation support system for contemporary dance e-learning. We developed a Body-part Motion Synthesis System (BMSS) that allows users to create choreographies by synthesizing body-part motions to increase the effect of learning contemporary dance choreography. Short created choreographies can be displayed as animation using 3DCG characters. The system targets students who are studying contemporary dance and is designed to promote the discovery learning of contemporary dance. We conducted a series of evaluation experiments for creating contemporary dance choreographies to verify the learning effectiveness of our system as a support system for discovery learning. As a consequence of experiments with 26 students who created contemporary dances, we verified that BMSS is a helpful creation training tool to discover new choreographic methods, new dance movements, and new awareness of their bodies.

In transcoding, it is well known that refinement of the motion vectors is critical to enhance the quality of transcoded video while significantly reducing transcoding complexity. This paper proposes a novel cost model to estimate the rate-distortion cost of motion vector composition in order to develop a reliable motion vector re-estimation method that has reasonable computation cost. Based on a statistical analysis of motion compensated prediction errors, we design a basic form of the proposed cost model as a function of distance from the optimal motion vector. Simulations with a transcoder employing the proposed cost model demonstrate a significant quality gain over representative video transcoding schemes with no complexity increase.

Negative electron affinity (NEA) surfaces can be formed by alternating supply of alkali metals (e.g. Cs, Rb, K) and oxygen on semiconductor surfaces. We have studied adsorption structures of Cs on an As-terminated (2×4) (001) GaAs surface using scanning tunneling microscopy (STM). We found that the initial adsorption of Cs atoms occurs around the step sites in the form of Cs clusters and that the size of clusters is reduced by successive exposure to O2, indicating that As-terminated (2×4) surfaces are relatively stable compared to Ga-terminated surfaces and are not broken by the Cs clusters adsorption.

To improve the performance of capsule endoscope, it is important to add location information to the image data obtained by the capsule endoscope. There is a disadvantage that a lot of existing localization techniques require to measure channel model parameters in advance. To avoid such a troublesome pre-measurement, this paper pays attention to capsule endoscope localization based on an electromagnetic imaging technology which can estimate not only the location but also the internal structure of a human body. However, the electromagnetic imaging with high resolution has huge computational complexity, which should prevent us from carrying out real-time localization. To ensure the accurate real-time localization system without pre-measured model parameters, we apply genetic algorithm (GA) into the electromagnetic imaging-based localization method. Furthermore, we evaluate the proposed GA-based method in terms of the simulation time and the location estimation accuracy compared to the conventional methods. In addition, we show that the proposed GA-based method can perform more accurately than the other conventional methods, and also, much less computational complexity of the proposed method can be accomplished than a greedy algorithm-based method.

The the depth-of-field limitation of our eyes causes out-of-focus blur in the retinal images. The blur dynamically changes whenever we change our gaze and accordingly the scene point we are looking at changes its depth. This paper proposes an image display that reproduces retinal out-of-focus blur by using a stereoscopic display and eye trackers. Its purpose is to provide the viewer with more realistic visual experiences than conventional (stereoscopic) displays. Unlike previous similar systems that track only one of the viewer's eyes to estimate the gaze depth, the proposed system tracks both eyes individually using two eye trackers and estimates the gaze depth from the convergence angle calculated by triangulation. This provides several advantages over existing schemes, such as being able to deal with scenes having multiple depths. We describe detailed implementations of the proposed system and show the results of an experiment conducted to examine its effectiveness. In the experiment, creating a scene having two depths using two LCD displays together with a half mirror, we examined how difficult it is for viewers to distinguish between the real scene and its virtual reproduction created by the proposed display system. The results of the experiment show the effectiveness of the proposed approach.

This paper describes recent progress of photonically-enabled systems for millimeter-wave and terahertz measurement applications. After briefly explaining signal generation schemes as a foundation of photonics-based approach, system configurations for specific applications are discussed. Then, practical demonstrations are presented, which include frequency-domain spectroscopy, phase-sensitive measurement, electric-field measurement, and 2D/3D imaging.

In this paper, an electromagnetic plane wave diffraction by finite number of loaded thick slits on infinitely long perfectly electric conductor (PEC) screen is analyzed. Here we formulate the problem by utilizing the Kobayashi Potential (KP) method, which is a kind of eigenfunction expansion method in terns of Weber-Schafheitlin discontinuous integrals. The multiple scattering contributions between the slits are analytically included in the formulation. The solution derived here may provide us with precise numerical result, so it may be considered as a reference solution to other numerical and approximate analyses.

We herein describe an autonomous peer discovery scheme for Device-to-Device (D2D) communications. With the increasing popularity of D2D communications, an efficient means of finding the neighboring node, i.e., peer discovery, is required. To this end, we propose a new autonomous peer discovery scheme that uses azimuth spread (AS), delay spread (DS), and shadow fading of the uplink pilot from each mobile station (MS). Given that AS, DS, and shadow fading are spatially correlated, nodes that have similar values must be neighbors. The proposed scheme filters out the MSs that are unlikely to be neighbors and uses the Kolmogorov-Smirnov (K-S) test to improve the accuracy of neighbor discovery. Unlike previous peer discovery schemes that incur additional signaling overheads, our proposal finds neighboring nodes by using the existing uplink pilot transmission from MSs such that neighboring peers can be found autonomously. Through analysis and simulation, we show that neighboring MSs can be found accurately with low latency.

Unsupervised spoken unit discovery or zero-source speech recognition is an emerging research topic which is important for spoken document analysis of languages or dialects with little human annotation. In this paper, we extend our earlier joint training framework for unsupervised learning of discrete density HMM to continuous density HMM (CDHMM) and apply it to spoken unit discovery. In the proposed recipe, we first cluster a group of Gaussians which then act as initializations to the joint training framework of nonnegative matrix factorization and semi-continuous density HMM (SCDHMM). In SCDHMM, all the hidden states share the same group of Gaussians but with different mixture weights. A CDHMM is subsequently constructed by tying the top-N activated Gaussians to each hidden state. Baum-Welch training is finally conducted to update the parameters of the Gaussians, mixture weights and HMM transition probabilities. Experiments were conducted on word discovery from TIDIGITS and phone discovery from TIMIT. For TIDIGITS, units were modeled by 10 states which turn out to be strongly related to words; while for TIMIT, units were modeled by 3 states which are likely to be phonemes.

This paper describes a hierarchical and cooperative transport system with demand responsive buses to improve service quality of public transport system in city area and its suburbs. To provide the demand responsive buses generally requires planning route and schedule called dial-a-ride problem. However, the problem complexity increases with the increasing of the number of requests. Therefore, we propose the hierarchical and cooperative transport system. Framework of the system can reduce scale of the problem by grouping customers. We have evaluated the proposed system on a static simulation and a dynamic microscopic simulation. The simulation result has shown the system could improve service quality by reducing customer's load. Moreover, the result of the dynamic simulation have provided the detailed features of the system.

In this paper, we propose SimCS (similarity based on contribution scores) to compute the similarity of scientific papers. For similarity computation, we exploit a notion of a contribution score that indicates how much a paper contributes to another paper citing it. Also, we consider the author dominance of papers in computing contribution scores. We perform extensive experiments with a real-world dataset to show the superiority of SimCS. In comparison with SimCC, the-state-of-the-art method, SimCS not only requires no extra parameter tuning but also shows higher accuracy in similarity computation.

This letter presents a method for active noise cancelation (ANC) for headphone application. The method improves the performance of ANC by deriving a flexible independent component analysis (ICA) algorithm in a hybrid structure combining feedforward and feedback configurations with correlation-based wind detection. The effectiveness of the method is demonstrated through simulation.

We realize homogenous luminance of the directional backlight for the time-division multiplexing autostereoscopic display using a convex lens array with the elemental lenses whose phase of placement in each row differs from one another. The validity of the proposed optical design is confirmed by a prototype system.